Coal mine drilling operation state while-drilling monitoring system and monitoring method

Abstract

The invention discloses a coal rock drilling operation state while-drilling monitoring system and method. The system is characterized in that a pressure and flow sensor I is mounted on a circulating hydraulic pipeline of a feeding oil cylinder; a pressure and flow sensor II is mounted on a circulating hydraulic pipeline of the hydraulic rotary motor; a displacement sensor is mounted on the feeding oil cylinder; the video collector is mounted above a coal wall of a construction roadway; the data acquisition and processing module is connected with each sensor; the video analysis module is connected with the video collector; the state determination module is connected with the data acquisition and processing module and the video analysis module. The method comprises the following steps: acquiring a multi-source drilling state signal and working image data; preprocessing the data; respectively acquiring state classification feature information and drill rod operation state image data, and inputting the state classification feature information and the drill rod operation state image data into a double-flow neural network to extract feature vectors; vector splicing is carried out through a Softmax gating fusion mechanism, and then a judgment result is output through a Softmax classifier. According to the method, the running state of the coal rock drilling can be efficiently and accurately judged.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent monitoring while drilling of coal mine underground detection drilling, pressure relief drilling, gas extraction drilling and the like, and specifically discloses a coal rock drilling operation state monitoring while drilling system and a monitoring method, which are mainly used for monitoring and analyzing the coal rock drilling operation state in the drilling construction process in the field of coal mine underground drilling. BACKGROUND

[0002] The monitoring while drilling operation is the key to realizing the determination of the coal rock drilling operation state, and the accurate identification of the unstable operation state is the key technology for correctly controlling the coal mine drilling machine to respond to the unstable state, which can effectively improve the production efficiency of the drilling machine and ensure the safe performance of the operation process. Under normal circumstances, the coal seam distribution of the underground coal mine has the characteristics of non-uniformity, randomness of coal rock hardness, and complexity of coal rock geological environment, so that the drilling operation has the uncertainty of drilling load. However, the existing coal rock drilling state analysis system has poor reliability in parameter identification, and the possibility of sticking accidents increases, which may cause the drill bit to fall off, the drill pipe to break, and the like, thereby damaging the drilling machine and even causing serious safety accidents.

[0003] Therefore, it is urgent to provide a monitoring system and a monitoring method capable of monitoring the coal rock drilling operation state, so that the coal mine drilling machine can efficiently and accurately determine the drilling operation state according to the change of the external load. SUMMARY

[0004] In view of the problems existing in the prior art, the present application provides a coal rock drilling operation state monitoring while drilling system and a monitoring method, which can collect multi-source monitoring data in the drilling construction process of the coal mine underground in real time and online and automatically process, can efficiently and accurately determine the operation state of the coal rock drilling based on the change of the external load, and is beneficial to greatly improving the safety factor of the drilling construction process; the method has high intelligence and loudness, and the implementation process is simple, which can efficiently and accurately determine the operation state of the drilling system, can significantly improve the drilling efficiency of the coal mine drilling machine, can solve the problems of poor response to unstable state characteristics of the coal mine drilling machine and insufficient in drilling state parameter identification in the prior art, and can effectively ensure the safety factor of the drilling machine in the coal wall drilling pressure relief, which is beneficial to better guarantee the safety of the pressure relief drilling, gas extraction and detection drilling operation.

[0005] In order to achieve the above purpose, the present application provides a coal rock drilling operation state monitoring while drilling system, which comprises an intelligent perception unit, a data acquisition and processing module, a video analysis module and a state determination module. The intelligent sensing unit includes a pressure sensor, a flow sensor, a pressure sensor, a flow sensor, a displacement sensor, and a video acquisition unit. Pressure sensor and flow sensor are both installed on the circulating hydraulic lines of the feed cylinder in the coal mine drilling rig, and are used to collect feed pressure and flow signals during the drilling process, respectively. Pressure sensor and flow sensor are both installed on the circulating hydraulic lines of the hydraulic rotary motor in the coal mine drilling rig, and are used to collect rotary pressure and flow signals during the drilling process, respectively. The displacement sensor is installed in the rodless chamber of the feed cylinder in the coal mine drilling rig, and is used to collect telescopic displacement signals within the feed cylinder. The video acquisition unit is installed above the coal face in the construction roadway, and is used to collect image data at the working position. The data acquisition and processing module is connected to pressure sensor one, flow sensor one, pressure sensor two, flow sensor two, and displacement sensor respectively. It is used to receive feed pressure signal, feed flow signal, slewing pressure signal, slewing flow signal, and extension displacement signal. It performs noise reduction processing on the received signals using minimum mean square error adaptive filtering technology to obtain multi-source drilling status signals. At the same time, it is used to classify the drilling status parameters of the multi-source drilling status signals to obtain status classification feature information. The video analysis module is connected to the video acquisition unit to receive image data from the working position and perform noise reduction processing on the image data at the working position to obtain noise-reduced working image data; at the same time, it classifies the working image data based on the RCNN deep learning model to obtain drill pipe running status image data. The status determination module is connected to the data acquisition and processing module and the video analysis module respectively. It is used to process the status classification feature information and drill pipe operation status image data based on the dual-stream neural network architecture to obtain and output the drilling operation status of the drilling rig.

[0006] As a preferred option, the video capture device is a high-speed industrial camera.

[0007] In this invention, pressure sensor one and flow sensor one, installed on the circulating hydraulic pipeline of the feed cylinder in the coal mine drilling rig, respectively collect feed pressure and feed flow signals during the drilling process. The feed pressure signal reflects the resistance between the drill bit and the coal / rock contact surface; a sudden increase in pressure may indicate changes in rock hardness or drill bit wear. Furthermore, the feed flow signal provides information on changes in feed flow rate. Simultaneously, pressure sensor two and flow sensor two, installed on the circulating hydraulic pipeline of the hydraulic rotary motor in the coal mine drilling rig, respectively collect rotary pressure and rotary flow signals during the drilling process. The rotary pressure signal indicates the drill bit rotation resistance; high pressure accompanied by low rotation speed may indicate drill bit jamming or rock fissure development. The rotary flow signal further provides information on changes in rotary flow rate. The addition of displacement sensors facilitates real-time sensing of feed displacement data. A video acquisition device facilitates the acquisition of working image data including the drill rod's operating status. Therefore, by collecting multiple signals, the drilling status can be comprehensively and accurately reflected. The data acquisition and processing module effectively denoises the signals collected by various sensors, resulting in cleaner monitoring signals. Furthermore, classifying the feature parameters of these signals yields multiple relevant key features for determining the drilling status, facilitating accurate subsequent assessments. Similarly, the video analysis module denoises the image data, producing cleaner images. Using an RCNN deep learning model, the image data is then classified, automatically, efficiently, and accurately obtaining feature vectors representing the drilling status. Finally, the status determination module integrates status classification feature information with drill pipe operation image data to obtain more accurate drilling status assessment results.

[0008] This system can collect multi-source monitoring data in real time during the drilling process in coal mines and process it automatically. It can efficiently and accurately determine the operating status of coal and rock boreholes based on changes in external load, which is conducive to significantly improving the safety factor of the drilling process. It is suitable for monitoring conditions during drilling, such as exploration boreholes, pressure relief boreholes, and gas extraction boreholes in coal mines.

[0009] This invention also provides a method for monitoring the operating status of coal and rock boreholes during drilling, employing a coal and rock borehole operating status monitoring system, comprising the following steps: Step 1: Obtain drilling data; During the pressure relief construction of coal and rock drilling, pressure sensor one and flow sensor one are used to collect the feed pressure signal and feed flow signal during the drilling process, respectively. Pressure sensor two and flow sensor two are used to collect the rotation pressure signal and rotation flow signal during the drilling process, respectively. Displacement sensor is used to collect the telescopic displacement signal in the feed cylinder. The feed pressure signal, feed flow signal, rotation pressure signal, rotation flow signal, and telescopic displacement signal are sent as multi-source drilling status signals to the data acquisition and processing module. The video acquisition device is used to collect image data at the working position and send the image data at the working position to the video analysis module. Step 2: Data preprocessing; The data acquisition and processing module uses the minimum mean square error adaptive filtering technique to denoise the multi-source drilling status signal, resulting in a denoised multi-source drilling status signal. At the same time, the video analysis module uses the image data at the working position to denoise the image data, resulting in a denoised working image data. Step 3: Feature extraction; C1: Obtain state classification feature information; C11: Using multi-feature integration analysis technology, features are extracted and fused from multi-source drilling status signals to construct a real-time feature vector set; C12: Construct an XGBoost deep learning model based on Bayesian optimization, and train the deep learning model using the historical feature vector set in the database to obtain a signal classification model after training. C13: The real-time feature vector set is used as input data and fed into the signal classification model to classify the drilling state parameters, obtain state classification feature information, and send it to the state judgment module. C2: Acquire image data of drill pipe operating status; The video analysis module uses an RCNN deep learning model to classify the image data at the working position to obtain the drill pipe operating status image data, which is then sent to the status judgment module. Step 4: Determine the running status and output the result; D1: Construct a two-stream neural network that simultaneously includes 1D-CNN branches and 3D-CNN branches using 1D-CNN and 3D-CNN; D2: Input the state classification feature information as input data into the 1D-CNN branch to extract the parameter depth feature vector; at the same time, input the drill pipe running status image data as input data into the 3D-CNN branch to extract the image depth feature vector. D3: The weights of the parameter depth feature vector and the image depth feature vector are dynamically exercised through the Softmax gated fusion mechanism, and normalized by the Softmax function. The parameter depth feature vector and the image depth feature vector are directly concatenated into a fused feature vector in the channel dimension. The fused feature vector is then input into the Softmax classifier for feature classification, and finally the result of the drilling operation status is output.

[0010] As a preferred option, the noise reduction process in step two is as follows: B1: Using a parameter-adjustable digital filter to process the original input signal The output signal of the adjustable digital filter is obtained by performing filtering according to formula (1). ; (1); In the formula, For a parameter-adjustable digital filter, the weight vector is... ; B2: Transform formula (1) to obtain formula (2); (2); B3: Obtained according to formula (3) Error signal of time-adjustable digital filter ; (3); In the formula, The desired output signal; B4: Based on the minimum mean square error criterion, construct the minimum mean square error according to formula (4). ; (4); B5: The weight vector is iteratively updated using the steepest descent method according to formula (5). ; (5); In the formula, Step size factor; The gradient of the statistical average of multiple squared error sequences. , The gradient of a single squared error sequence; B6: Equation (5) is transformed to obtain equation (6); (6); B7: By executing formulas (5) and (6) It gradually converges to the optimal solution, and eventually, makes mean square value Minimum.

[0011] As a preferred option, in step C11 of step three, the feature extraction process is as follows: C11-1: For pressure signals, according to formula (7), the short-time Fourier transform is used. The spectrum is obtained; for the displacement signal, the velocity curve is obtained by calculating the forward displacement difference according to formula (8). For the flow signal, the time-varying function of the hydraulic rotary motor is obtained according to formula (9). ; (7); In the formula, The original signal; For window functions; It is a complex exponential base; (8); (9); In the formula, For instantaneous flow, It is the instantaneous angular velocity. Leakage coefficient, The instantaneous pressure difference between the two ends of the motor. It is a compressible flow rate; C11-2: According to formula (10), the correlation between the feature parameters is represented by the Spearman correlation coefficient. ; (10); In the formula, The amount of data for the feature parameters; It represents the difference in the order of the two feature parameters.

[0012] As a preferred option, in step C12 of step three, the process of optimizing the XGBoost deep learning model based on Bayesian principles is as follows: C12-1: For those by An XGBoost consisting of 3 basic trees has m drilling state classifications, namely C1, C2...Cm. The initial XGBoost objective function is defined according to formula (11). The complexity of a decision tree is defined according to formula (12). ; (11); In the formula, The number of samples; For the first The actual value of each sample; For the front The predicted value of each tree, with the loss function being: , The complexity of a tree; (12); In the formula, T is the number of leaf nodes, and γ is a parameter that controls the complexity of the tree. This represents the value of the j-th leaf node in the tree; C12-2: Define the parameter space for optimization: Select the learning rate, the maximum depth of the tree, the minimum sample weights required for leaf nodes, and the minimum loss function reduction required when splitting nodes as optional parameters of the model. Specifically, the learning rate is defined to be within [0.01, 0.3], the maximum depth of the tree to be within [3, 10], the minimum sample weights required for leaf nodes to be within [1, 10], and the minimum loss function reduction required when splitting nodes to be within [0, 1]. C12-3: Initialization; Randomly select several initial sample points, evaluate their performance, and use these points to initialize the Gaussian process model, where the Gaussian process model is shown in formula (13); (13); In the formula, It is a mean function, used to describe how the average value of a function changes with the input; The covariance function is used to describe the correlation between function values ​​at different input points. C12-4: Iterative optimization; Select the next experimental point, evaluate the objective function on the sample points, validate it using cross-validation, and update the Gaussian over-optimized model. The new test results are added to the Gaussian process model, and the parameters of the model are continuously updated until the predetermined number of iterations is reached or the condition is met; the cross-validation method is shown in formula (14); (14); In the formula, Training algorithm using hyperparameter θ This is the training data of the i-th fold. This is the verification data for the i-th fold; C12-5: Selecting the optimal parameters: Select the best-performing set of parameters from all evaluated combinations as the final result and save it to obtain the optimal signal classification model.

[0013] As a preferred embodiment, in step three, C13, the drilling condition classification process is as follows: Based on the maximum input torque of the drilling system With rotational load torque The relationship between the two, and the drilling speed of the drill string. Critical rotational speed with stuck drill The relationships between them are classified; among them, , This refers to the inlet pressure of the hydraulic motor. This refers to the outlet pressure of the hydraulic motor. Hydraulic motor displacement, The mechanical efficiency of the hydraulic motor; , For the resistance of coal and rock to torque, For frictional torque, The diameter of the drill bit. Torsional strength of coal and rock The rate of increase of friction, For drilling depth, Drill pipe radius; where, , This refers to the rotational speed of the rotary motor. The transmission ratio is... This represents the overall transmission efficiency. The flow rate of the hydraulic rotary motor. The volumetric efficiency of the hydraulic rotary motor; , The drilling speed of the drill bit is The corresponding excitation frequency; The specific classification process is as follows: when or If the load is stable, the drilling system can automatically return to stability and enter normal drilling conditions, outputting state one. when or If the load is within the fluctuating operating condition, the drilling system will experience slight stuck drill bit, and output state two. when or At this point, the rotary pressure rises rapidly, the drilling system becomes stuck, and the output status is three.

[0014] As a preferred option, in step C2 of step three, the process of classifying the state of the image data at the working location based on the RCNN deep learning model is as follows: C21: Collect historical drill pipe operating status image information data, preprocess it, classify the processed results, and establish a sample library; C22: Input the video image information of the drill pipe's operating status into the RCNN deep learning model to generate multiple candidate regions that may contain the target object; C23: Input each candidate region into a pre-trained CNN model to extract a fixed-length depth feature vector; C24: Using depth feature vectors as input data, a multi-class SVM classifier is trained. Each class uses a binary SVM, and the final output is the class probability. The class probabilities correspond to the stable operating state, the drilling fluctuation state, and the stuck drill state, respectively. The stable operating state indicates that the drill rod rotates at a constant speed and the drilling is smooth. The drilling fluctuation state indicates that the drill rod swings slightly, the rotation speed and axial displacement speed fluctuate, and the backfill flow rate slows down. The stuck drill state indicates that the drill rod has no axial displacement and high-frequency vibration occurs at the joint, or the drill rod is obviously bent, deviated, or broken.

[0015] As a preferred embodiment, in step four, D2, the extraction process of the parameter depth feature vector is as shown in D21, and the extraction process of the image depth feature vector is as shown in D22. D21: The state classification feature information processed by the data acquisition and processing module is used as input data and input to the 1D-CNN branch of the state determination module. After three layers of convolution, the feature map is obtained. Finally, adaptive global average pooling is used to obtain a 128-dimensional temporal global feature vector, which is used as the parameter depth feature vector. D22: The drill pipe running status image data processed by the video analysis module is used as input data and input to the 3D-CNN branch of the status determination module. First, the image features are initially extracted through 3*7*7 convolution kernels and 1*2*2 max pooling. Then, it goes through three cascaded P3D residual stages. Each P3D residual stage first uses 1*3*3 spatial convolution and then 3*1*1 temporal convolution to further extract image features. Finally, global spatiotemporal pooling is used to obtain a 512-dimensional feature vector, which is used as the image depth feature vector.

[0016] As a preferred option, in step four, D3, the Softmax gated fusion mechanism is as shown in formula (15); (15); In the formula, , Given two feature vectors, This is the concatenation of two vectors. This is the weight matrix. This is the bias vector. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the drilling operation status monitoring system for coal and rock boreholes in this invention. Figure 2 This is a flowchart of the method for monitoring the operating status of coal and rock boreholes during drilling in this invention; Figure 3 This is a basic schematic diagram of the adaptive filtering principle in this invention; Figure 4This is a flowchart of the Bayesian-optimized XGBoost method in the monitoring method of this invention; Figure 5 This is a flowchart illustrating the data processing using a two-stream neural network architecture in this invention. Detailed Implementation

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] like Figures 1 to 5 As shown, the present invention provides a coal and rock borehole operation status monitoring system, including an intelligent sensing unit, a data acquisition and processing module, a video analysis module and a status determination module; The intelligent sensing unit includes a pressure sensor, a flow sensor, a pressure sensor, a flow sensor, a displacement sensor, and a video acquisition unit. Pressure sensor and flow sensor are both installed on the circulating hydraulic lines of the feed cylinder in the coal mine drilling rig, and are used to collect feed pressure and flow signals during the drilling process, respectively. Pressure sensor and flow sensor are both installed on the circulating hydraulic lines of the hydraulic rotary motor in the coal mine drilling rig, and are used to collect rotary pressure and flow signals during the drilling process, respectively. The displacement sensor is installed in the rodless chamber of the feed cylinder in the coal mine drilling rig, and is used to collect telescopic displacement signals within the feed cylinder. The video acquisition unit is installed above the coal face in the construction roadway, and is used to collect image data at the working position. The data acquisition and processing module is connected to pressure sensor one, flow sensor one, pressure sensor two, flow sensor two, and displacement sensor respectively. It is used to receive feed pressure signal, feed flow signal, slewing pressure signal, slewing flow signal, and extension displacement signal. It performs noise reduction processing on the received signals using minimum mean square error adaptive filtering technology to obtain multi-source drilling status signals. At the same time, it is used to classify the drilling status parameters of the multi-source drilling status signals to obtain status classification feature information. The video analysis module is connected to the video acquisition unit to receive image data from the working position and perform noise reduction processing on the image data at the working position to obtain noise-reduced working image data; at the same time, it classifies the working image data based on the RCNN deep learning model to obtain drill pipe running status image data. The status determination module is connected to the data acquisition and processing module and the video analysis module respectively. It is used to process the status classification feature information and drill pipe operation status image data based on the dual-stream neural network architecture to obtain and output the drilling operation status of the drilling rig.

[0020] In this invention, pressure sensor one and flow sensor one, installed on the circulating hydraulic pipeline of the feed cylinder in the coal mine drilling rig, respectively collect feed pressure and feed flow signals during the drilling process. The feed pressure signal reflects the resistance between the drill bit and the coal / rock contact surface; a sudden increase in pressure may indicate changes in rock hardness or drill bit wear. Furthermore, the feed flow signal provides information on changes in feed flow rate. Simultaneously, pressure sensor two and flow sensor two, installed on the circulating hydraulic pipeline of the hydraulic rotary motor in the coal mine drilling rig, respectively collect rotary pressure and rotary flow signals during the drilling process. The rotary pressure signal indicates the drill bit rotation resistance; high pressure accompanied by low rotation speed may indicate drill bit jamming or rock fissure development. The rotary flow signal further provides information on changes in rotary flow rate. The addition of displacement sensors facilitates real-time sensing of feed displacement data. A video acquisition device facilitates the acquisition of working image data including the drill rod's operating status. Therefore, by collecting multiple signals, the drilling status can be comprehensively and accurately reflected. The data acquisition and processing module effectively denoises the signals collected by various sensors, resulting in cleaner monitoring signals. Furthermore, classifying the feature parameters of these signals yields multiple relevant key features for determining the drilling status, facilitating accurate subsequent assessments. Similarly, the video analysis module denoises the image data, producing cleaner images. Using an RCNN deep learning model, the image data is then classified, automatically, efficiently, and accurately obtaining feature vectors representing the drilling status. Finally, the status determination module integrates status classification feature information with drill pipe operation image data to obtain more accurate drilling status assessment results.

[0021] This system can collect multi-source monitoring data in real time during the drilling process in coal mines and process it automatically. It can efficiently and accurately determine the operating status of coal and rock boreholes based on changes in external load, which is conducive to significantly improving the safety factor of the drilling process. It is suitable for monitoring conditions during drilling, such as exploration boreholes, pressure relief boreholes, and gas extraction boreholes in coal mines.

[0022] As a preferred option, the video capture device is a high-speed industrial camera.

[0023] This invention also provides a method for monitoring the operating status of coal and rock boreholes during drilling, employing a coal and rock borehole operating status monitoring system, comprising the following steps: Step 1: Obtain drilling data; During the pressure relief construction of coal and rock drilling, pressure sensor one and flow sensor one are used to collect the feed pressure signal and feed flow signal during the drilling process, respectively. Pressure sensor two and flow sensor two are used to collect the rotation pressure signal and rotation flow signal during the drilling process, respectively. Displacement sensor is used to collect the telescopic displacement signal in the feed cylinder. The feed pressure signal, feed flow signal, rotation pressure signal, rotation flow signal, and telescopic displacement signal are sent as multi-source drilling status signals to the data acquisition and processing module. The video acquisition device is used to collect image data at the working position and send the image data at the working position to the video analysis module. Step 2: Data preprocessing; The data acquisition and processing module uses the minimum mean square error adaptive filtering technique to denoise the multi-source drilling status signal, resulting in a denoised multi-source drilling status signal. At the same time, the video analysis module uses the image data at the working position to denoise the image data, resulting in a denoised working image data. Specifically, in the denoising process, the minimum mean square error adaptive filtering technique is used to decompose the acquired original signal so as to minimize the mean square error between the output signal and the desired signal.

[0024] Step 3: Feature extraction; C1: Obtain state classification feature information; C11: Using multi-feature integration analysis technology, features are extracted and fused from multi-source drilling status signals to construct a real-time feature vector set; C12: Construct an XGBoost deep learning model based on Bayesian optimization, and train the deep learning model using the historical feature vector set in the database to obtain a signal classification model after training. C13: The real-time feature vector set is used as input data and fed into the signal classification model to classify the drilling state parameters, obtain state classification feature information, and send it to the state judgment module. C2: Acquire image data of drill pipe operating status; The video analysis module uses an RCNN deep learning model to classify the image data at the working position to obtain the drill pipe operating status image data, which is then sent to the status judgment module. Step 4: Determine the running status and output the result; D1: Construct a two-stream neural network that simultaneously includes 1D-CNN branches and 3D-CNN branches using 1D-CNN and 3D-CNN; D2: Input the state classification feature information as input data into the 1D-CNN branch to extract the parameter depth feature vector; at the same time, input the drill pipe running status image data as input data into the 3D-CNN branch to extract the image depth feature vector. D3: The weights of the parameter depth feature vector and the image depth feature vector are dynamically exercised through the Softmax gated fusion mechanism, and normalized by the Softmax function. The parameter depth feature vector and the image depth feature vector are directly concatenated into a fused feature vector in the channel dimension. The fused feature vector is then input into the Softmax classifier for feature classification, and finally the result of the drilling operation status is output.

[0025] As a preferred option, the noise reduction process in step two is as follows: B1: Using a parameter-adjustable digital filter to process the original input signal The output signal of the adjustable digital filter is obtained by performing filtering according to formula (1). ; (1); In the formula, For a parameter-adjustable digital filter, the weight vector is... ; B2: Transform formula (1) to obtain formula (2); (2); B3: Obtained according to formula (3) Error signal of time-adjustable digital filter ; (3); In the formula, The desired output signal; B4: Based on the minimum mean square error criterion, construct the minimum mean square error according to formula (4). ; (4); B5: To find the optimal weight vector, the steepest descent method is used to iteratively update the weight vector according to formula (5). ; (5); In the formula, Step size factor; The core idea of ​​minimum mean square error (MSE) is to statistically average the gradient of multiple squared error sequences, replacing the mean squared error with the squared error. , The gradient of a single squared error sequence; B6: Equation (5) is transformed to obtain equation (6); (6); B7: By executing formulas (5) and (6) It gradually converges to the optimal solution, and eventually, makes mean square value Minimum.

[0026] As a preferred option, in step C11 of step three, the feature extraction process is as follows: C11-1: For pressure signals, according to formula (7), the short-time Fourier transform is used. The spectrum is obtained; for the displacement signal, the velocity curve is obtained by calculating the forward displacement difference according to formula (8). For the flow signal, the time-varying function of the hydraulic rotary motor is obtained according to formula (9). ; (7); In the formula, The original signal; For window functions; It is a complex exponential base; (8); (9); In the formula, For instantaneous flow, It is the instantaneous angular velocity. Leakage coefficient, The instantaneous pressure difference between the two ends of the motor. It is a compressible flow rate; C11-2: According to formula (10), the correlation between the feature parameters is represented by the Spearman correlation coefficient. ; (10); In the formula, The amount of data for the feature parameters; It represents the difference in the order of the two feature parameters.

[0027] As a preferred option, in step C12 of step three, the process of optimizing the XGBoost deep learning model based on Bayesian principles is as follows: C12-1: For those by An XGBoost consisting of 3 basic trees has m drilling state classifications, namely C1, C2...Cm. The initial XGBoost objective function is defined according to formula (11). The complexity of a decision tree is defined according to formula (12). ; (11); In the formula, The number of samples; For the first The actual value of each sample; For the front The predicted value of each tree, with the loss function being: , The complexity of a tree; (12); In the formula, T is the number of leaf nodes, and γ is a parameter that controls the complexity of the tree. This represents the value of the j-th leaf node in the tree; C12-2: Define the parameter space for optimization: Select the learning rate, the maximum depth of the tree, the minimum sample weights required for leaf nodes, and the minimum loss function reduction required when splitting nodes as optional parameters of the model. Specifically, the learning rate is defined to be within [0.01, 0.3], the maximum depth of the tree to be within [3, 10], the minimum sample weights required for leaf nodes to be within [1, 10], and the minimum loss function reduction required when splitting nodes to be within [0, 1]. C12-3: Initialization; Randomly select several initial sample points, evaluate their performance, and use these points to initialize the Gaussian process model. The Gaussian process is a commonly used probabilistic model in Bayesian optimization, derived from the mean function. and kernel function Completely determined, wherein the Gaussian process model is shown in formula (13); (13); In the formula, It is a mean function, used to describe how the average value of a function changes with the input; The covariance function is used to describe the correlation between function values ​​at different input points. C12-4: Iterative optimization; Select the next experimental point, evaluate the objective function on the sample points, validate it using cross-validation, and update the Gaussian over-optimized model. The new test results are added to the Gaussian process model, and the parameters of the model are continuously updated until the predetermined number of iterations is reached or the condition is met; the cross-validation method is shown in formula (14); (14); In the formula, Training algorithm using hyperparameter θ This is the training data of the i-th fold. This is the verification data for the i-th fold; C12-5: Selecting the optimal parameters: Select the best-performing set of parameters from all evaluated combinations as the final result and save it to obtain the optimal signal classification model.

[0028] As a preferred option, in step three, C13, the drilling status classification process is as follows: Based on the maximum input torque of the drilling system With rotational load torque The relationship between the drilling tools and the drilling speed Critical rotational speed with stuck drill The relationships between them are classified; among them, , This refers to the inlet pressure of the hydraulic motor. This refers to the outlet pressure of the hydraulic motor. Hydraulic motor displacement, The mechanical efficiency of the hydraulic motor; , For the resistance of coal and rock to torque, For frictional torque, The diameter of the drill bit. Torsional strength of coal and rock The rate of increase of friction, For drilling depth, Drill pipe radius; where, , This refers to the rotational speed of the rotary motor. The transmission ratio is... This represents the overall transmission efficiency. The flow rate of the hydraulic rotary motor. The volumetric efficiency of the hydraulic rotary motor; , The drilling speed of the drill bit is The corresponding excitation frequency; The specific classification process is as follows: when or If the load is stable, the drilling system can automatically return to stability and enter normal drilling conditions, outputting state one. when or If the load is within the fluctuating operating condition, the drilling system will experience slight stuck drill bit, and output state two. when or At this point, the rotary pressure rises rapidly, the drilling system becomes stuck, and the output status is three.

[0029] As a preferred option, in step C2 of step three, the process of classifying the state of the image data at the working location based on the RCNN deep learning model is as follows: C21: To verify the effectiveness of the RCNN algorithm in drilling status classification, historical drill pipe operating status image information data were collected, preprocessed, and the processed results were classified to establish a sample library. C22: Input the video image information of the drill pipe's operating status into the RCNN deep learning model to generate multiple candidate regions that may contain the target object; C23: Input each candidate region into a pre-trained CNN model to extract a fixed-length depth feature vector; C24: Using depth feature vectors as input data, a multi-class SVM classifier is trained. Each class uses a binary SVM, and the final output is the class probability. The class probabilities correspond to the stable operating state, the drilling fluctuation state, and the stuck drill state, respectively. The stable operating state indicates that the drill rod rotates at a constant speed and the drilling is smooth. The drilling fluctuation state indicates that the drill rod swings slightly, the rotation speed and axial displacement speed fluctuate, and the backfill flow rate slows down. The stuck drill state indicates that the drill rod has no axial displacement and high-frequency vibration occurs at the joint, or the drill rod is obviously bent, deviated, or broken.

[0030] As a preferred embodiment, in step four, D2, the extraction process of the parameter depth feature vector is as shown in D21, and the extraction process of the image depth feature vector is as shown in D22. D21: The state classification feature information processed by the data acquisition and processing module is used as input data and input to the 1D-CNN branch of the state determination module. After three layers of convolution, the feature map is obtained. Finally, adaptive global average pooling is used to obtain a 128-dimensional temporal global feature vector, which is used as the parameter depth feature vector. D22: The drill pipe running status image data processed by the video analysis module is used as input data and input to the 3D-CNN branch of the status determination module. First, the image features are initially extracted through 3*7*7 convolution kernels and 1*2*2 max pooling. Then, it goes through three cascaded P3D residual stages. Each P3D residual stage first uses 1*3*3 spatial convolution and then 3*1*1 temporal convolution to further extract image features. Finally, global spatiotemporal pooling is used to obtain a 512-dimensional feature vector, which is used as the image depth feature vector.

[0031] As a preferred option, in step four, D3, the Softmax gated fusion mechanism is as shown in formula (15); (15); In the formula, , Given two feature vectors, This is the concatenation of two vectors. This is the weight matrix. This is the bias vector.

[0032] This invention provides a method for monitoring the drilling status of coal and rock boreholes during drilling. First, multiple sources of drilling parameters are collected through various sensors during the drilling process, and image data from the working position is simultaneously acquired via a video acquisition device, enabling comprehensive and accurate determination of the drilling status. Second, the signals from the sensors are denoised using a minimum mean square error adaptive filtering technique, effectively removing irrelevant interference information and significantly improving the signal-to-noise ratio of the monitoring signals. This effectively preserves key features such as drill bit resistance and rock hardness, providing a clean data source for drilling status diagnosis. Denoising the image data using a video analysis module effectively eliminates interference noise and restores drilling contour details, avoiding edge blurring and facilitating the acquisition of more accurate drill rod operating status image data. Furthermore, classifying pressure, displacement, and flow characteristic signals using a Bayesian-optimized XGBoost deep learning model yields multiple correlated key feature information for accurate and efficient judgment of the drilling status, facilitating subsequent accurate determination of the drilling status. Simultaneously, classifying the image data using an RCNN deep learning model automatically, efficiently, and accurately obtains feature vectors representing the drilling status. Finally, the state determination module based on the dual-stream neural network architecture fuses state classification feature information and drill pipe operation status image data through the Softmax gated fusion mechanism. The fused feature vector is then input into the Softmax classifier for feature classification, which can significantly improve the accuracy of drilling state determination through collaborative analysis.

[0033] This method boasts high intelligence and simple implementation. By analyzing and processing drilling operation parameters and drill rod operation image information, it can efficiently and accurately determine the operating status of the drilling system, significantly improving the drilling efficiency of coal mine drilling rigs. It can also be effectively applied to the monitoring and analysis of the operating status of coal and rock boreholes during the drilling process in underground coal mine exploration. This method solves the problems of poor characteristics of coal mine drilling rigs in dealing with instability and insufficient identification of drilling status parameters in existing technologies. It can effectively ensure the safety factor of coal mine drilling rigs in relieving pressure during drilling in coal walls, which is conducive to better preventing rockburst accidents.

Claims

1. A monitoring system for the operating status of coal and rock boreholes during drilling, characterized in that, It includes an intelligent sensing unit, a data acquisition and processing module, a video analysis module, and a status determination module; The intelligent sensing unit includes a pressure sensor, a flow sensor, a pressure sensor, a flow sensor, a displacement sensor, and a video acquisition unit. Pressure sensor and flow sensor are both installed on the circulating hydraulic lines of the feed cylinder in the coal mine drilling rig, and are used to collect feed pressure and flow signals during the drilling process, respectively. Pressure sensor and flow sensor are both installed on the circulating hydraulic lines of the hydraulic rotary motor in the coal mine drilling rig, and are used to collect rotary pressure and flow signals during the drilling process, respectively. The displacement sensor is installed in the rodless chamber of the feed cylinder in the coal mine drilling rig, and is used to collect telescopic displacement signals within the feed cylinder. The video acquisition unit is installed above the coal face in the construction roadway, and is used to collect image data at the working position. The data acquisition and processing module is connected to pressure sensor one, flow sensor one, pressure sensor two, flow sensor two, and displacement sensor respectively. It is used to receive feed pressure signal, feed flow signal, slewing pressure signal, slewing flow signal, and extension displacement signal. It performs noise reduction processing on the received signals using minimum mean square error adaptive filtering technology to obtain multi-source drilling status signals. At the same time, it is used to classify the drilling status parameters of the multi-source drilling status signals to obtain status classification feature information. The video analysis module is connected to the video acquisition unit and is used to receive image data at the acquisition work position and perform noise reduction processing on the image data at the work position to obtain noise-reduced working image data. Meanwhile, the working image data is classified based on the RCNN deep learning model to obtain drill pipe operating status image data; The status determination module is connected to the data acquisition and processing module and the video analysis module respectively. It is used to process the status classification feature information and drill pipe operation status image data based on the dual-stream neural network architecture to obtain and output the drilling operation status of the drilling rig.

2. The coal and rock borehole operation status monitoring system according to claim 1, characterized in that, The video capture device is a high-speed industrial camera.

3. A method for monitoring the operating status of coal and rock boreholes during drilling, employing a coal and rock borehole operating status monitoring system as described in claim 1 or 2, characterized in that, Includes the following steps: Step 1: Obtain drilling data; During the pressure relief construction of coal and rock drilling, pressure sensor one and flow sensor one are used to collect the feed pressure signal and feed flow signal during the drilling process, respectively. Pressure sensor two and flow sensor two are used to collect the rotation pressure signal and rotation flow signal during the drilling process, respectively. Displacement sensor is used to collect the telescopic displacement signal in the feed cylinder. The feed pressure signal, feed flow signal, rotation pressure signal, rotation flow signal, and telescopic displacement signal are sent as multi-source drilling status signals to the data acquisition and processing module. The video acquisition device is used to collect image data at the working position and send the image data at the working position to the video analysis module. Step 2: Data preprocessing; The data acquisition and processing module uses the minimum mean square error adaptive filtering technique to denoise the multi-source drilling status signal, resulting in a denoised multi-source drilling status signal. At the same time, the video analysis module uses the image data at the working position to denoise the image data, resulting in a denoised working image data. Step 3: Feature extraction; C1: Obtain state classification feature information; C11: Using multi-feature integration analysis technology, features are extracted and fused from multi-source drilling status signals to construct a real-time feature vector set; C12: Construct an XGBoost deep learning model based on Bayesian optimization, and train the deep learning model using the historical feature vector set in the database to obtain a signal classification model after training. C13: The real-time feature vector set is used as input data and fed into the signal classification model to classify the drilling state parameters, obtain state classification feature information, and send it to the state judgment module. C2: Acquire image data of drill pipe operating status; The video analysis module uses an RCNN deep learning model to classify the image data at the working position to obtain the drill pipe operating status image data, which is then sent to the status judgment module. Step 4: Determine the running status and output the result; D1: Construct a two-stream neural network that simultaneously includes 1D-CNN branches and 3D-CNN branches using 1D-CNN and 3D-CNN; D2: Input the state classification feature information as input data into the 1D-CNN branch to extract the parameter depth feature vector; at the same time, input the drill pipe running status image data as input data into the 3D-CNN branch to extract the image depth feature vector. D3: The weights of the parameter depth feature vector and the image depth feature vector are dynamically exercised through the Softmax gated fusion mechanism, and normalized by the Softmax function. The parameter depth feature vector and the image depth feature vector are directly concatenated into a fused feature vector in the channel dimension. The fused feature vector is then input into the Softmax classifier for feature classification, and finally the result of the drilling operation status is output.

4. The method for monitoring the operating status of coal and rock boreholes during drilling according to claim 3, characterized in that, In step two, the noise reduction process is as follows: B1: Using a parameter-adjustable digital filter to process the original input signal The output signal of the adjustable digital filter is obtained by performing filtering according to formula (1). ; （1）； In the formula, For a parameter-adjustable digital filter, the weight vector is... ; B2: Transform formula (1) to obtain formula (2); （2）； B3: Obtained according to formula (3) Error signal of time-adjustable digital filter ; （3）； In the formula, The desired output signal; B4: Based on the minimum mean square error criterion, construct the minimum mean square error according to formula (4). ; （4）； B5: The weight vector is iteratively updated using the steepest descent method according to formula (5). ; （5）； In the formula, Step size factor; The gradient of the statistical average of multiple squared error sequences. , The gradient of a single squared error sequence; B6: Equation (5) is transformed to obtain equation (6); （6）； B7: By executing formulas (5) and (6) It gradually converges to the optimal solution, and eventually, makes mean square value Minimum.

5. The method for monitoring the operating status of coal and rock boreholes during drilling according to claim 3, characterized in that, In step C11 of step three, the feature extraction process is as follows: C11-1: For pressure signals, according to formula (7), the short-time Fourier transform is used. The spectrum is obtained; for the displacement signal, the velocity curve is obtained by calculating the forward displacement difference according to formula (8). ; For the flow signal, the time-varying function of the hydraulic rotary motor is obtained according to formula (9). ; （7）； In the formula, The original signal; For window functions; It is a complex exponential base; （8）； （9）； In the formula, For instantaneous flow, It is the instantaneous angular velocity. Leakage coefficient, The instantaneous pressure difference between the two ends of the motor. It is a compressible flow rate; C11-2: According to formula (10), the correlation between the feature parameters is represented by the Spearman correlation coefficient. ; （10）； In the formula, The amount of data for the feature parameters; It represents the difference in the order of the two feature parameters.

6. The method for monitoring the operating status of coal and rock boreholes during drilling according to claim 3, characterized in that, In step C12 of step three, the process of optimizing the XGBoost deep learning model based on Bayesian principles is as follows: C12-1: For those by An XGBoost consisting of 3 basic trees has m drilling state classifications, namely C1, C2...Cm. The initial XGBoost objective function is defined according to formula (11). The complexity of a decision tree is defined according to formula (12). ; （11）； In the formula, The number of samples; For the first The actual value of each sample; For the front The predicted value of each tree, with the loss function being: , The complexity of a tree; （12）； In the formula, T is the number of leaf nodes, and γ is a parameter that controls the complexity of the tree. This represents the value of the j-th leaf node in the tree; C12-2: Define the parameter space for optimization: Select the learning rate, the maximum depth of the tree, the minimum sample weights required for leaf nodes, and the minimum loss function reduction required when splitting a node as optional parameters of the model. Specifically, the learning rate is defined to be within [0.01, 0.3], the maximum depth of the tree to be within [3, 10], the minimum sample weights required for leaf nodes to be within [1, 10], and the minimum loss function reduction required when splitting a node to be within [0, 1]. C12-3: Initialization; Randomly select several initial sample points, evaluate their performance, and use these points to initialize the Gaussian process model, where the Gaussian process model is shown in formula (13); （13）； In the formula, It is a mean function, used to describe how the average value of a function changes with the input; The covariance function is used to describe the correlation between function values ​​at different input points. C12-4: Iterative optimization; Select the next experimental point, evaluate the objective function on the sample points, validate it using cross-validation, and update the Gaussian over-optimized model. The new test results are added to the Gaussian process model, and the parameters of the model are continuously updated until the predetermined number of iterations is reached or the condition is met; the cross-validation method is shown in formula (14); （14）； In the formula, Training algorithm using hyperparameter θ This is the training data of the i-th fold. This is the verification data for the i-th fold; C12-5: Selecting the optimal parameters: Select the best-performing set of parameters from all evaluated combinations as the final result and save it to obtain the optimal signal classification model.

7. The method for monitoring the operating status of coal and rock boreholes during drilling according to claim 6, characterized in that, In step three, C13, the drilling status classification process is as follows: Based on the maximum input torque of the drilling system With rotational load torque The relationship between the drilling tools and the drilling speed Critical rotational speed with stuck drill The relationships between them are categorized; among them, , This refers to the inlet pressure of the hydraulic motor. This refers to the outlet pressure of the hydraulic motor. Hydraulic motor displacement, The mechanical efficiency of the hydraulic motor; , For the resistance of coal and rock to torque, For frictional torque, The diameter of the drill bit. Torsional strength of coal and rock The rate of increase of friction, For drilling depth, Drill pipe radius; where, , This refers to the rotational speed of the rotary motor. The transmission ratio is... This represents the overall transmission efficiency. The flow rate of the hydraulic rotary motor. The volumetric efficiency of the hydraulic rotary motor; , The drilling speed of the drill bit is The corresponding excitation frequency; The specific classification process is as follows: when or If the load is stable, the drilling system can automatically return to stability and enter normal drilling conditions, outputting state one. when or If the load is within the fluctuating operating condition, the drilling system will experience slight stuck drill bit, and output state two. when or At this point, the rotary pressure rises rapidly, the drilling system becomes stuck, and the output status is three.

8. The method for monitoring the operating status of coal and rock boreholes during drilling according to claim 6, characterized in that, In step C2 of step three, the process of classifying the state of the image data at the working location based on the RCNN deep learning model is as follows: C21: Collect historical drill pipe operating status image information data, preprocess it, classify the processed results, and establish a sample library; C22: Input the video image information of the drill pipe's operating status into the RCNN deep learning model to generate multiple candidate regions that may contain the target object; C23: Input each candidate region into a pre-trained CNN model to extract a fixed-length depth feature vector; C24: Using depth feature vectors as input data, a multi-class SVM classifier is trained. Each class uses a binary SVM, and the final output is the class probability. The class probabilities correspond to the stable operating state, the drilling fluctuation state, and the stuck drill state, respectively. The stable operating state indicates that the drill rod rotates at a constant speed and the drilling is smooth. The drilling fluctuation state indicates that the drill rod swings slightly, the rotation speed and axial displacement speed fluctuate, and the backfill flow rate slows down. The stuck drill state indicates that the drill rod has no axial displacement and high-frequency vibration occurs at the joint, or the drill rod is obviously bent, deviated, or broken.

9. The method for monitoring the operating status of coal and rock boreholes during drilling according to claim 5, characterized in that, In step four, D2, the extraction process of the parameter depth feature vector is shown in D21, and the extraction process of the image depth feature vector is shown in D22. D21: The state classification feature information processed by the data acquisition and processing module is used as input data and input to the 1D-CNN branch of the state determination module. After three layers of convolution, the feature map is obtained. Finally, adaptive global average pooling is used to obtain a 128-dimensional temporal global feature vector, which is used as the parameter depth feature vector. D22: The drill pipe running status image data processed by the video analysis module is used as input data and input to the 3D-CNN branch of the status determination module. First, the image features are initially extracted through 3*7*7 convolution kernels and 1*2*2 max pooling. Then, it goes through three cascaded P3D residual stages. Each P3D residual stage first uses 1*3*3 spatial convolution and then 3*1*1 temporal convolution to further extract image features. Finally, global spatiotemporal pooling is used to obtain a 512-dimensional feature vector, which is used as the image depth feature vector.

10. The method for monitoring the operating status of coal and rock boreholes during drilling according to claim 5, characterized in that, In step four, D3, the Softmax gated fusion mechanism is shown in formula (15); （15）； In the formula, , Given two feature vectors, This is the concatenation of two vectors. This is the weight matrix. This is the bias vector.

Images