Intelligent identification method for underground coal mine rotary drilling stratum lithology

Abstract

The invention discloses a coal mine underground rotary drilling formation lithology intelligent identification method, which is based on an intelligent drilling machine, a CSTT system and a mining drilling visual measuring instrument, and is characterized in that the intelligent drilling machine and the CSTT system cooperate in a drilling construction process to automatically collect a drilling cutting image sample; remeasuring the drill holes by a drill hole multi-parameter measuring device to obtain stratum lithology data, and forming a model training data set; a set learning method is adopted to train and form a drilling cuttings lithology identification model based on the drilling cuttings image; in the normal construction stage, the trained model is integrated into a drilling machine control system, collected drilling cuttings images are detected in real time, and real-time recognition of the lithology of the drilling cuttings and the proportions of the drilling cuttings with the different lithology is achieved. According to the method, the consistency of the drilling cutting image acquisition environment, acquisition time and acquisition process is ensured, the lithology identification precision is improved, and automatic calculation of different lithology drilling cutting proportions is realized; and data and equipment support is provided for intelligent sensing of stratum information and a self-adaptive drilling technology in the drilling construction process.

Description

Technical Field

[0001] The present invention belongs to the field of formation information perception during the construction of underground coal mine drilling holes, and relates to an intelligent identification method for formation lithology of underground coal mine rotary drilling based on automatic sampling and analysis of drill cuttings. Background Art

[0002] With the continuous development of drilling equipment and drilling construction technology, underground drilling construction in coal mines has gradually been automated and is moving towards a more intelligent direction. Intelligent sensing technology for the drilling construction process, represented by coal-rock identification technology, is a difficult problem that must be overcome to achieve intelligent drilling construction. Lithology identification technology can not only provide data support for the development of adaptive drilling technology, but also be used to obtain more accurate resource endowment information, ensure that the drilling covers the target extraction area, and ensure the gas extraction effect.

[0003] Traditional technology relies on drillers to observe changes in the shape and color of drill cuttings to determine the lithology of the construction stratum during the drilling process, and manually record the drilling depth at the intersection of coal and rock, and use the drilling trajectory re-measurement data to calculate the coal seam position. This method is affected by the driller's working status and knowledge level, and has low accuracy. These recorded data will be used to guide the subsequent drilling trajectory design, affecting the effect of gas extraction or disaster management.

[0004] With the continuous development of intelligent drilling technology, researchers have conducted a lot of research on the problem of rapid identification of rock properties during drilling construction. Intelligent real-time identification of coal and rock based on construction parameters, vibration of drilling rigs or drilling tools, sound and other information generated during the drilling construction process can reduce subjective human judgment errors and improve identification accuracy and efficiency. However, due to the differences in formation pressure, gas and water storage conditions between different mines, the drilling construction data will also change. The generalization ability of the resulting lithology identification model is usually not high and can only be used in specific research areas. Most of the related research on lithology identification based on directly collected rock or core images and image recognition methods remains in the laboratory research stage, failing to achieve real-time identification of coal and rock during the drilling construction process and unable to guide drilling construction in a timely manner. In recent years, although artificial intelligence has made great progress in image recognition, its application in underground coal mine cuttings analysis to obtain accurate lithology information still faces continuous challenges.

[0005] These challenges include:

[0006] 1. The manual sampling method cannot accurately control the sampling process, resulting in a spatial and temporal mismatch between the acquired lithology information and the sampling location information;

[0007] To achieve the precise spatio-temporal correspondence between cuttings images and sampling locations is very important for accurately identifying formation information. The ideal cuttings sampling interval is to conduct cuttings sampling at regular intervals along the borehole trajectory, forming a uniformly distributed cuttings lithology information along the borehole trajectory direction. However, in the rotary drilling borehole construction method, it is difficult to use a probe tube to obtain the accurate trajectory of the borehole during the borehole construction process. Differences in formation lithology, formation pressure, formation rock mechanical properties, etc. will all cause changes in the borehole construction speed. Therefore, using the traditional method of sampling at regular intervals of time cannot achieve the precise spatio-temporal correspondence between cuttings images and sampling locations.

[0008] 2. The harsh environment underground in coal mines and the lack of necessary standardized image acquisition processes will seriously affect the quality of the collected images and reduce the lithology identification accuracy.

[0009] The harsh underground environment in coal mines, such as water mist, dust, and darkness, will have a greater impact on the quality of cuttings images. Different image sampling devices, image acquisition parameters, and color calibration will also affect the image recognition accuracy.

[0010] 3. Existing cuttings image recognition methods will compress high-definition cuttings images to meet the input requirements of machine learning models, resulting in the loss of a large amount of effective information.

[0011] The original image data collected by the image sensor is usually a high-resolution cuttings image. When using traditional methods to train a pre-trained network using transfer learning, the image needs to be adjusted to the required size, such as 224 pixels × 224 pixels. This adjustment may lead to the loss of a large amount of valuable information. Since cuttings are usually a mixture of coal and rock particles, understanding the proportion of coal and rock particles in the cuttings is more meaningful for obtaining the accurate position of the coal-rock interface and guiding the drilling operation.

[0012] 4. Existing methods cannot obtain the proportion of coal cuttings and rock cuttings in the cuttings sample, resulting in a reduction in formation identification accuracy. For example: when the borehole just enters the coal seam from the rock formation, the cuttings collected may mainly consist of rock cuttings, with only a small number of coal cuttings particles. Traditional coal-rock identification methods based on cuttings images usually identify the entire cuttings image as rock. However, at this time, the borehole has entered the coal seam, resulting in a deviation between the identification result and the actual position of the coal-rock interface.

[0013] 5. There is a lack of data communication between the lithology identification device and the drill rig, and it is impossible to guide the borehole construction in real time.

[0014] These factors further hinder the practical application of machine learning-based lithology identification methods in coal mines. Summary of the Invention

[0015] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method for intelligently identifying stratum lithology during rotary drilling in coal mines, so as to realize intelligent perception of stratum lithology information during drilling construction in coal mines.

[0016] In order to solve the above technical problems, the present invention adopts the following technical solutions to achieve the above problems:

[0017] A method for intelligently identifying lithology of rotary drilling strata in underground coal mines, the method comprising:

[0018] Step 1, model training phase: demonstration drilling trajectory design, system joint debugging, drilling construction and automatic collection of drill cuttings images, drilling retest, drill cuttings image preprocessing, image enhancement and establishment of training data set, training and optimization of individual classification models, and construction of ensemble learning models;

[0019] Step 2, deployment and implementation phase: integration of ensemble learning models, automatic collection of drill cuttings images, image preprocessing, lithology identification and result output.

[0020] The present invention also includes the following technical features:

[0021] Specifically, the trajectory design of the demonstration drilling covers typical strata in the mining area and potential geological structures that need to be identified in the future to collect sufficient data to train the lithology identification machine learning model.

[0022] Specifically, in the joint debugging of the system, the system includes an intelligent drilling rig, a CSTT system, and a mining drilling visualization measuring instrument;

[0023] The intelligent drilling rig includes: a drilling system, an angle adjustment system, a control system, a crawler walking system, an automatic loading and unloading system for drill rods, a pump station, and a controller;

[0024] The CSTT system includes an orifice device and a host, the host includes a cover plate assembly, an image acquisition device, a weighing device, a drill cuttings conveying device, a frame assembly, and a slag conveying pipe; the cover plate assembly, the image acquisition device, the weighing device, and the drill cuttings conveying device are arranged in the frame assembly from top to bottom; the cover plate assembly includes a cover plate and a slag inlet thereon; the orifice device is arranged at the borehole opening, one end of which is connected to the borehole opening, and the other end is connected to the slag inlet on the cover plate assembly through the slag conveying pipe; the image acquisition device includes an image collector, a sampling tray, and a sampling mechanical arm, and the sampling mechanical arm can drive the sampling tray to move between the drill cuttings collection point below the slag inlet and the image collection point below the image collector;

[0025] The mining drilling visualization measuring instrument can efficiently obtain underground drilling high-definition video, three-dimensional trajectory and coal-rock characteristics to quickly identify the stratum lithology in the direction of the tunnel trajectory;

[0026] The joint debugging of the system includes: installing the intelligent drill rig and the CSTT system in the drill field, completing the pipeline connection and conducting joint debugging; adjusting the drill rig to the correct inclination, azimuth, and opening position according to the borehole design, and stabilizing the drill rig body; installing an opening bit on the drill rig, completing the borehole opening, and installing the outer casing pipe; installing the orifice device of the CSTT system at the borehole orifice, inserting the orifice pipe of the orifice device into the outer casing pipe of the borehole, and sealing the annulus; connecting the orifice device and the host of the CSTT system with a slag discharge pipe; connecting each actuator of the CSTT system to the control valve with corresponding functions on the drill rig through hydraulic hoses; debugging to ensure the normal operation and data transmission of the drill rig and the CSTT system.

[0027] Specifically, the automatic collection of borehole construction and drill cuttings images includes:

[0028] Sort out the automatic construction process of the intelligent drill rig borehole and the drill cuttings image collection process of the CSTT system, and divide the entire borehole construction process into 15 drill rig operating states; obtain the drill rig operating state in real time through sensors, the system starts to read the position of the drill rig power head relative to the fuselage, and the drill rig control unit sends a start sampling command to the drill cuttings image processing unit; the hydraulic system controls the drive cylinder of the sampling manipulator to slowly retract, and sends the sampling tray to the sampling position; during this period, the pressure sensor monitors the pressure of the drive cylinder, and when the hydraulic pressure reaches 20 MPa, the system considers that the drive cylinder is in the target position; the sampling tray flips, and the cuttings are poured into the weighing device; after sampling, control the sampling manipulator to move the sampling tray to the image capture position; during this process, level the drill cuttings in the sampling tray; the drill cuttings sample stays at the image capture position to drain the moisture; subsequently, the image collector starts to capture the drill cuttings image and transmits it to the drill cuttings image processing unit in the drill rig control.

[0029] Specifically, the borehole remeasurement includes: after the borehole construction is completed, use a mine borehole visualization measuring instrument to remeasure the borehole to obtain accurate borehole trajectory information and formation lithology change information along the borehole trajectory direction; combine the logging data of the borehole and the known geological information of the mine to form the true lithology data at each sampling point.

[0030] The preprocessing of the drill cuttings image includes: screening of the original image, establishment of the model training data set, and establishment of the model test data set.

[0031] The screening of the original image: According to the borehole remeasurement data, eliminate the drill cuttings image samples collected from the interface between two different lithology formations.

[0032] The establishment of the model training data set includes:

[0033] Initial cropping: Crop the original drill cuttings image to remove invalid information such as the edge of the sampling container.

[0034] Secondary cropping: According to the requirements of the selected image recognition deep learning model for the input image;

[0035] Numbering: Number the cuttings image blocks obtained after secondary cropping;

[0036] Lithology annotation: Annotate the lithology of each image according to the true experimental data of the cuttings image acquisition points obtained previously, and then divide the image blocks into different groups according to different lithologies to form a model training data set;

[0037] The establishment of the model test data set includes: randomly selecting 80% of the data in the model training data set for the training of the machine learning model, and 20% of the data forms the model test data set for testing the accuracy change during the training process of the machine learning model.

[0038] Specifically, the establishment of the image enhancement and training data set includes:

[0039] Adopt the runtime enhancement technology. When reading the original training image, transform or combine-transform it with a given probability of 80%; then add the transformed image to the current training batch for model training; after the training ends, the transformed image will not be saved;

[0040] The training and optimization of the individual classification model:

[0041] Use the model training data set and the model test data set to test these common models and select the deep learning pre-trained models with high recognition accuracy for cuttings images. Then, use transfer learning to perform intensive training on these pre-trained models; by selecting different optimization algorithms, changing the model structure, adjusting the initial learning rate and validation frequency, etc., optimize the recognition accuracy of the individual classification model and select the model with both recognition accuracy and speed.

[0042] Specifically, the construction of the ensemble learning model:

[0043] Based on the soft voting ensemble learning method, use multiple individual classification models that have been intensively trained previously to establish an ensemble learning model for lithology recognition based on cuttings images; by adjusting the weights of each model, form the corresponding relationship between the weight sum and the recognition accuracy of the ensemble model, so as to obtain the optimal weight combination and construct the optimal ensemble learning model; use the model test data set to test, optimize and evaluate the model to ensure that it can accurately classify the cuttings image samples; after passing the test, complete the training work of the lithology recognition model.

[0044] Specifically, the integration of the set learning model: In the deployment and implementation phase, the model is integrated into the control system of the intelligent drilling rig; the drill cuttings image data is read from the drilling rig controller in real time and identified; the identification results are then transmitted and stored in the storage unit of the intelligent drilling rig controller and simultaneously displayed on the screen of the drilling rig's control system.

[0045] The automatic acquisition of the drill cuttings images: During the drilling operation, the intelligent drilling rig and the CSTT system work in coordination. For each drill pipe constructed, one drill cuttings image is collected and transmitted to the drilling rig controller.

[0046] Specifically, the image preprocessing: The preprocessing method for normal construction data: After the model is trained, optimized, and integrated into the drilling rig system, the preprocessing method for the original drill cuttings images is slightly different from that in the model training phase; first, the collected image data is cropped, and then each image block is randomly rotated to generate multiple new image blocks; these newly generated image blocks, together with the original image blocks, constitute the lithology identification data set for this specific image block.

[0047] Specifically, the lithology identification and result output: After the identification of a single drill cuttings image is completed, the number and identification accuracy of the drill cuttings image blocks identified as different lithologies are counted. By selecting the category with the highest weighted average probability in the lithology identification data set, the lithology identification result of each image block is determined; according to the proportion of the image blocks with the same lithology type in the total number of image blocks, the distribution proportion of this lithology in the drill cuttings is determined.

[0048] Compared with the prior art, the present invention has the following technical effects:

[0049] The present invention innovatively develops the automatic completion of drill cuttings sampling and lithology analysis for the hardware system; innovatively proposes a preprocessing method for the broken coal and rock images, making it possible to automatically calculate the lithology proportion; the standard image acquisition process, constant image acquisition environment, and unified image acquisition timing ensure the image quality; multiple innovative measures reduce the occurrence of drill cuttings mixing; real-time data communication, and the lithology identification results provide online guidance for the drilling operation.

[0050] The present invention ensures the consistency of the drill cuttings image acquisition environment, acquisition timing, and acquisition process, reducing the influence of the mixing of drill cuttings from different strata on the formation identification accuracy. And through the innovative preprocessing method for the coal and rock broken samples, the lithology identification accuracy is improved, and the automatic calculation of the proportion of drill cuttings of different lithologies is realized. It provides data and equipment support for the intelligent perception of formation information during the drilling operation and the research and development of the adaptive drilling technology. Brief Description of the Drawings

[0051] Figure 1 It is a system composition, construction layout, and method flow chart of the present invention.

[0052] Figure 2 This is the structural composition of the automatic sampling, weighing and transportation device for drill cuttings of the present invention.

[0053] Figure 3 This is the schematic diagram of the image acquisition system of the present invention.

[0054] Figure 4 This is the schematic diagram of the drill cuttings image collector of the present invention.

[0055] Figure 5 This is the schematic diagram of the automatic sampling process of drill cuttings of the present invention.

[0056] Figure 6 This is the schematic diagram of the data processing flow of the present invention.

[0057] Figure 7 This is the image preprocessing method in the model training stage of the present invention: (a) Image cropping and numbering; (b) Lithology marking of the image.

[0058] Figure 8 This is the image preprocessing method under the normal construction state of the present invention.

[0059] The meanings of the various labels in the figure are as follows:

[0060] 1. Cover plate assembly, 2. Image acquisition device, 3. Weighing device, 4. Drill cuttings conveying device, 5. Frame assembly, 1.1 Inlet for slag, 1.2 Cover plate, 2.1.1 Intrinsically safe camera, 2.1.2 Mounting flange, 2.1.3 LED light source, 2.1.4 Light guide plate, 2.1.5 Light homogenizing plate. Detailed implementation method

[0061] In view of the deficiencies existing in the prior art, integrating the experience and achievements of long-term engagement in related professions, through painstaking research and design, the present invention proposes an intelligent identification method for the lithology of the underground rotary drilling formation in coal mines based on the automatic sampling and analysis of drill cuttings, and develops a complete set of equipment to support the implementation of this method. It ensures the consistency of the drill cuttings image acquisition environment, acquisition timing, and acquisition process, reduces the influence of the mixing of drill cuttings from different strata on the formation identification accuracy, and improves the lithology identification accuracy through an innovative preprocessing method for coal and rock broken samples, realizing the automatic calculation of the proportion of drill cuttings of different lithologies. It provides data and equipment support for the intelligent perception of formation information during the drilling construction process and the research and development of adaptive drilling technology.

[0062] The present invention discloses an intelligent identification method for stratum lithology of rotary drilling in coal mines based on automatic sampling and analysis of drill cuttings, which is implemented based on an intelligent drilling rig, an automatic sampling, weighing and transportation system for drill cuttings, and a visual measuring instrument for drilling holes for mining. During the drilling process, the intelligent drilling rig and the automatic sampling, weighing and transportation system for drill cuttings cooperate with each other to automatically collect drill cuttings image samples. In the lithology identification model training stage based on deep learning, the drilling hole is repeatedly measured by a multi-parameter measuring device to obtain stratum lithology data, and a model training data set is formed. A drill cutting lithology identification model based on drill cuttings images is trained by a set learning method. During the normal construction stage, the trained model is integrated into the drilling rig control system, and the drill cuttings images collected by the automatic sampling, weighing and transportation system for drill cuttings are detected in real time, so as to realize the real-time identification of the drill cuttings lithology and the proportion of drill cuttings of different lithologies during the drilling process. The method adopts a variety of measures to ensure the consistency of the drill cuttings image acquisition environment, acquisition timing, and acquisition process, and reduce the influence of the mixing of drill cuttings from different strata on the stratum identification accuracy. Through innovative coal and rock crushing sample preprocessing methods, the accuracy of lithology identification has been improved, and the automatic calculation of the proportion of drill cuttings of different lithologies has been realized, providing data and equipment support for the intelligent perception of stratum information and the research and development of adaptive drilling technology during drilling construction.

[0063] Specific embodiments of the present invention are given below. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent changes made on the basis of the technical solution of this application fall within the protection scope of the present invention.

[0064] Embodiment 1:

[0065] like Figures 1 to 8 As shown, this embodiment provides a method for intelligently identifying lithology of rotary drilling formations in coal mines. The method is implemented based on an intelligent drilling rig, a cuttings automatic sampling, weighing and transportation system, i.e., a CSTT system, and a visual measuring instrument for drilling holes in a mine; the method includes:

[0066] Step 1, model training phase: demonstration drilling trajectory design, system joint debugging, drilling construction and automatic collection of drill cuttings images, drilling retest, drill cuttings image preprocessing, image enhancement and establishment of training data set, training and optimization of individual classification models, and construction of ensemble learning models;

[0067] Step 2, deployment and implementation phase: integration of ensemble learning models, automatic collection of drill cuttings images, image preprocessing, lithology identification and result output.

[0068] The model training phase specifically includes the following:

[0069] 1.1 Demonstration drilling trajectory design

[0070] During the model training phase, several demonstration drill holes need to be constructed to collect sufficient data for training the lithology identification machine learning model. The trajectory planning of the demonstration drill holes should cover the typical strata in the mining area and the potential geological structures to be identified in the future.

[0071] 1.2 Joint commissioning of the system

[0072] In the joint commissioning of the system, the system includes an intelligent drill rig, a CSTT system, and a mine borehole visualization measurement instrument;

[0073] The intelligent drill rig includes: a drilling system, an angle adjustment system, a control system, a crawler walking system, a drill pipe automatic loading and unloading system, a pump station, a controller, etc.; multiple different types of sensors are installed on the drill rig to obtain the operating status of the drill rig. For example, a displacement sensor is installed between the rotary table and the feed body of the drill rig to monitor the position of the rotary table relative to the feed body in real time; a rotational speed sensor is integrated in the rotary table to monitor the rotational speed of the rotary table; pressure transmitters are installed at key nodes of the hydraulic system to detect the pressure changes at key nodes of the hydraulic system, such as: the driving pressure of the feed system, the driving pressure of the rotary system, and the return oil pressure. A controller is also integrated on the drill rig, and the data processing capacity of the control is required to simultaneously meet the usage requirements of image recognition and borehole construction data analysis, and also take into account the data collection and storage functions. The intelligent drill rig leaves an interface for the CSTT system, and the interface includes a data communication interface and a hydraulic drive interface; among them, the data communication interface realizes the transmission of drill cuttings image data, sampling instruction data, and command execution status data through CAN bus communication; the hydraulic drive interface is responsible for providing power to each actuator of the CSTT system. The data of each sensor, control system, and controller on the intelligent drill rig is communicated in real time through the CAN bus.

[0074] The CSTT system includes an orifice device and a host computer. The host computer includes a cover plate assembly 1, an image acquisition device 2, a weighing device 3, a drill cuttings conveying device 4, a frame assembly 5, and a slag discharge pipe. The cover plate assembly 1, the image acquisition device 2, the weighing device 3, and the drill cuttings conveying device 4 are arranged in the frame assembly 5 from top to bottom. The cover plate assembly 1 includes a cover plate 1.2 and a slag inlet 1.1 thereon. The orifice device is arranged at the drilling orifice, one end of which is connected to the drilling orifice, and the other end is connected to the slag inlet on the cover plate assembly through the slag discharge pipe. The image acquisition device 2 includes an image acquisition unit, a sampling tray, and a sampling robotic arm. The sampling robotic arm can drive the sampling tray to move between a drill cuttings collection point below the slag inlet and an image acquisition point below the image acquisition unit. More specifically, the image acquisition device 2 further includes a frame, a driving oil cylinder, a sampling tray and a flipping device, a flipping trigger plate, and a scraper. The weighing device 3 is suspended from the frame assembly 5 by tensile and compressive sensors arranged around it, and it is provided with a slag discharge door. The drill cuttings conveying device 4 can convey the drill cuttings discharged from the slag discharge door to a designated position. The CSTT system further includes a flow meter, a pumping station, a valve group, and a core controller. The pumping station provides energy for the whole system, and the valve group controls the actions of each actuator in the system under the control of the core controller. The flow meter is installed in the drilling fluid pipeline to monitor the flow rate of the drilling fluid flowing into the borehole.

[0075] Specifically, in the CSTT system, the image acquisition unit is installed at a corner of the frame. One end of the sampling robotic arm is arranged at another corner of the frame through a pin shaft. The other end of the sampling robotic arm is installed with the sampling tray and the flipping device. The driving oil cylinder is installed under the frame and connected to the sampling robotic arm to drive the sampling robotic arm to rotate horizontally around the pin shaft. When the driving oil cylinder is fully extended, the sampling robotic arm can rotate to make the sampling tray in the sampling tray and the flipping device located directly below the image acquisition unit. When the driving oil cylinder is fully retracted, the sampling tray can be moved directly below the slag inlet, realizing the movement of the sampling tray between the drill cuttings collection point and the image acquisition point. The image acquisition unit adopts an intrinsically safe design, meeting the requirements for use in coal mine underground. The image acquisition unit of the present invention includes an intrinsically safe camera, a mounting flange, an LED light source, a light guide plate, and a light homogenizing plate. The image acquisition unit creates a relatively enclosed environment between the intrinsically safe camera and the sampling tray and provides independent lighting to eliminate the influence of the surrounding environment on image acquisition. In addition, the intrinsically safe camera uses the same shooting parameters for each shot, such as shooting environment, shooting angle, camera settings, and exposure time, etc., further ensuring the consistency of the drill cuttings image sampling environment.

[0076] The sampling tray and the flipping device include a sampling tray, a tray mounting base plate, a pin, a double torsion spring, a base I, a base II, a torsion spring, a flipping trigger rod, a cover plate II, and a cover plate I; the sampling tray is disc-shaped and located at the hollow position of the tray mounting base plate. A baffle is welded to one end of the sampling tray, and the baffle contacts the tray mounting base plate, so that the tray can only flip to the side without the baffle; shafts are respectively arranged on both sides of the sampling tray. The left shaft is installed between the base I and the cover plate I on the tray mounting base plate and can rotate around the inner hole between the two. The pin is fixed on the left shaft by threads, and a double torsion spring is installed between the base I, the cover plate I, and the pin to provide a rotary reset force for the sampling tray; the right shaft is installed between the base II and the cover plate II on the tray mounting base plate and can rotate around the inner hole between the two. A quarter-circular boss is provided at the end of the right shaft; the flipping trigger rod is installed at the end of the right shaft. The flipping trigger rod is L-shaped and a semi-circular boss is machined at the corresponding end of the quarter-circular boss; the semi-circular boss on the flipping trigger rod meshes with the quarter-circular boss at the end of the right shaft, so that when the flipping trigger rod rotates 90 degrees clockwise, the sampling tray will rotate together; when the flipping trigger rod rotates 90 degrees counterclockwise, the sampling tray will not rotate together; a torsion spring is also fixed on the right shaft, and the elastic force of the torsion spring is less than the elastic force of the double torsion spring to provide a reset force for the flipping trigger rod, so that the L end of the L-shaped flipping trigger rod is in the vertical state at the initial position, and at the same time, the semi-circular boss and the quarter-circular boss are in contact.

[0077] A flipping trigger plate and a scraper are respectively arranged on the movement tracks of the sampling tray and the flipping device; the flipping trigger plate and the scraper are installed on the frame; the scraper can scrape the drill cuttings in the sampling tray flat after the sampling tray finishes collecting the drill cuttings; the flipping trigger plate can, after completing the image sampling, push the flipping trigger rod to rotate 90 degrees clockwise to pour the drill cuttings in the sampling tray into the weighing device below.

[0078] When the sampling tray finishes collecting drill cuttings and needs to be moved to the image acquisition position, it will pass by the scraper and the flipping trigger plate in sequence. The scraper levels the drill cuttings in the tray. The flipping trigger plate only contacts the flipping trigger rod and pushes the flipping trigger rod to rotate counterclockwise by 90 degrees. At this time, the sampling tray will not rotate together. When the flipping trigger rod passes through the flipping trigger plate, under the action of the restoring force of the torsion spring, the L end of the L-shaped flipping trigger rod returns to the vertical state. The robotic arm sends the collected drill cuttings to the image sampling position. After completing the image sampling, when the sampling tray needs to be moved to the drill cuttings sampling position, it will pass by the flipping trigger plate and the scraper in sequence. The flipping trigger plate will contact the flipping trigger rod and push the flipping trigger rod to rotate clockwise by 90 degrees. At this time, the sampling tray will rotate together and pour the drill cuttings in the tray into the weighing device below. When the flipping trigger rod passes through the flipping trigger plate, under the action of the restoring force of the torsion spring, the L end of the L-shaped flipping trigger rod returns to the vertical state. The robotic arm sends the sampling tray to the drill cuttings collection position. Before each sampling, the sampling tray and the flipping device will empty the original drill cuttings in the sampling tray, preventing the mixing of drill cuttings from different strata during the sampling process and affecting the collection accuracy of drill cuttings samples.

[0079] The weighing device includes a tension and compression sensor, a slag discharge driving oil cylinder, a slag discharge door, and a box body. The bottom surface cross-section of the box body is trapezoidal, and the inclined sides of the trapezoid are respectively provided with slag discharge doors, and the slag discharge doors can be opened and closed under the drive of the slag discharge driving oil cylinder.

[0080] The drill cuttings conveying device includes a welded box body, a water spraying port, a plunger suction pump I, a plunger suction pump II, a switching oil cylinder, a stirring impeller, a hydraulic motor, a slag discharge manifold, a slag outlet, a double-port flange plate, and a manifold flange;

[0081] The welded box body is an open-top container. One side of the container is sloped, and two groups of water spraying ports are provided at the top of the slope. The lower end of the slope is the main slag storage area of the box body, and two groups of stirring impellers are respectively provided at both ends, which can rotate under the drive of the hydraulic motor to stir the slag water evenly. A slag discharge manifold is provided in the middle of the slag storage area. The end of the slag discharge manifold is sealed with the double-port flange plate through the manifold flange and can switch the inner through hole between the plunger suction pump I and the plunger suction pump II under the drive of the switching oil cylinder. The other end of the manifold flange is provided with a slag outlet for connecting with the underground coal mine slag discharge pipeline.

[0082] The orifice device includes an orifice pipe, a slag collection device, a gas drainage port, and an orifice slag outlet. The orifice pipe is located at the front end of the device and is tubular, with an outer diameter slightly smaller than the drilling diameter and an inner diameter slightly larger than the outer diameter of the drill bit. The slag collection device is connected to the end of the orifice pipe through a flange. A gas drainage port is provided at the top of the slag collection device and is connected to the underground negative pressure drainage pipe through a pipeline during construction. A slag outlet is provided at the bottom of the slag collection device and is connected to the slag inlet on the cover plate assembly through a slag conveying pipe. The end of the slag collection device is sealed with the outer end of the drill pipe through a rubber cup to prevent gas leakage.

[0083] More specifically, the orifice device is installed at the orifice of the drill hole, and drill cuttings are collected through the annular space between the orifice pipe inserted into the hole and the drill pipe. The drill cuttings are conveyed through the slag conveying pipe to the slag inlet on the cover plate assembly and flow into the weighing device below. A tension and compression sensor is arranged on the weighing device to record the change information of the weight of the drill cuttings. A drill cuttings image acquisition device is arranged between the weighing device and the cover plate assembly, and a hydraulic cylinder can drive a robotic arm and a sampling tray at its end to automatically collect drill cuttings samples. A large number of small holes with a diameter of 5 mm are arranged at the bottom of the sampling tray, and the powdery drill cuttings will flow out of the tray along with the drilling fluid. An image collector is installed above the tray to collect drill cuttings images. After the drill cuttings in the weighing device reach a certain weight, the slag discharge port below the device will automatically open to discharge the drill cuttings into the drill cuttings conveying device below. A water spraying mechanism, a stirring mechanism and a pumping device are arranged in the conveying device, which can mix the drill cuttings and water evenly and convey them to the designated position through the pumping device.

[0084] In the CSTT system, data communication interfaces and hydraulic drive interfaces for rapid integration with the intelligent drill are reserved. The interfaces are used to cooperate with the corresponding interfaces of the intelligent drill to complete the transmission of drill cuttings image data, sampling instruction data, command execution status data and provide power for each actuator in the system. The CSTT system can be deeply integrated and work collaboratively with the intelligent drill. The drill cuttings image information collected by the CSTT system is transmitted to the drill controller in real time through the CAN bus communication protocol. The electromagnetic proportional multi-way valve of the intelligent drill is responsible for providing power for each actuator of the CSTT system.

[0085] The CSTT system integrates an image acquisition system, as Figure 3 shown. The image acquisition system includes an image collector, a sampling tray and a sampling robotic arm, and also includes a frame, a driving oil cylinder, a sampling tray and a flipping device, a flipping trigger plate, a scraper, and an image processing unit in the drill controller; as Figure 4 shown, the image collector includes an intrinsically safe camera 2.1.1, a mounting flange 2.1.2, an LED light source 2.1.3, a light guide plate 2.1.4 and a light homogenizing plate 2.1.5; the image collector creates a relatively enclosed environment between the intrinsically safe camera 2.1.1 and the sampling tray and provides independent lighting to eliminate the influence of the surrounding environment on image acquisition. In addition, the intrinsically safe camera 2.1.1 uses the same shooting parameters for each shot, such as shooting environment, shooting angle, camera settings and exposure time, etc., further ensuring the consistency of the drill cuttings image sampling environment.

[0086] The CSTT system and the intelligent drill share the same controller. The controller is divided into a drill cuttings image processing unit, a drilling construction parameter processing unit and a drill control unit, which are respectively responsible for the storage, preprocessing, lithology analysis of drill cuttings images and drilling construction parameters and the control of each automated action of the drill.

[0087] The mine drilling visualization measurement instrument adopts a micro multi-parameter integrated acquisition scheme, which can efficiently obtain high-definition videos, three-dimensional trajectories and coal and rock properties (natural gamma) of underground drill holes, and can help technicians quickly identify the formation lithology in the roadway trajectory direction;

[0088] The joint commissioning of the system includes: In this stage, the intelligent drill rig and the CSTT system need to be installed in the drill yard, the pipeline connection is completed and the joint commissioning is carried out. The drill yard layout and pipeline connection are as Figure 1 shown. Adjust the drill rig in place according to the inclination angle, azimuth angle and opening position of the drill hole design, use the stabilization system of the drill rig to level the drill rig body, and stabilize the drill rig. Install the opening bit on the drill rig, complete the drill hole opening and install the outer hole mouth pipe. Install the orifice device of the CSTT system at the drill hole orifice, insert the orifice pipe of the orifice device into the outer hole mouth pipe of the drill hole, and use materials such as cloth bags to seal the annulus between the orifice pipe of the orifice device and the outer hole mouth pipe. Place the dust collection box as close to the drill hole as possible, and use a chain to firmly fix the orifice device on the coal wall. Place the main unit of the CSTT system below the drill hole, close to the drill hole. Connect the orifice device and the main unit of the CSTT system with a slag discharge pipe. Connect each actuator of the CSTT system to the control valve with corresponding functions on the drill rig through a hydraulic hose. Connect the cable of the intrinsically safe camera to the reserved interface of the drill rig controller through an intrinsically safe aviation plug using an explosion-proof cable. The commissioning ensures that the actions and data transmission of the drill rig and the CSTT system are normal.

[0089] Before the drill hole construction starts, the CSTT system performs a self-check procedure to ensure smooth data communication, all actuators are in the initial position, the sensor data is zeroed, and it waits for the sampling instruction.

[0090] Self-check process:

[0091] (1) Data communication detection: The drill rig control unit sends a handshake command to the CSTT system control unit. After the CSTT system control unit receives the handshake command, it replies to the command. If the drill rig control unit receives this reply, it means that the data transmission is normal, and the screen of the controller shows "Data communication normal". On the contrary, if this reply is not received, it means that the data communication is abnormal, and the screen of the controller shows "Data communication abnormal";

[0092] (2) Manipulator detection: The sampling manipulator slowly retracts and then extends, and detects whether the pressure of the manipulator drive oil cylinder is ≥20 MPa during the retraction and extension processes. If both can, it outputs "Manipulator drive pressure normal";

[0093] (3) Sampling tray detection: When the robotic arm fully extends and the driving oil cylinder pressure ≥ 20 MPa, the image acquisition device captures an image and transmits it to the drill rig controller. This image is compared with the image of the empty sampling tray. If the difference is less than 30%, "Sampling tray normal" is output; otherwise, "Sampling tray abnormal" is output;

[0094] (4) Weighing device detection: The empty weight of the weighing device is 25 Kg. Check whether the sum M0 of the values of the four tension and compression sensors is less than 30 Kg. If it is less than 30 Kg, store M0 = 30 Kg and output "Weighing device zeroed"; if it is not less than 30 Kg, output "Weighing device abnormal";

[0095] (5) Drill cuttings conveying device detection: When the mixing device and the pumping device are running, check whether their driving pressures are less than 20 Mpa respectively. If less, output "Drill cuttings conveying device normal"; otherwise, output "Drill cuttings conveying device abnormal".

[0096] 1.3 Automatic acquisition of drilling construction and drill cuttings images

[0097] Through the deep integration of the CSTT system and the intelligent drill rig control system (including the integration of their electro-hydraulic control systems, the integration of the sampling process and the drilling construction process), the invention realizes the standardization of the drill cuttings sampling timing and the precise control of the sample sampling position.

[0098] The specific method is as follows: Sort out the automatic drilling construction process of the intelligent drill rig and the drill cuttings image acquisition process of the CSTT system. Divide the entire drilling construction process into 15 drill rig operating states as follows:

[0099] Table 1 Intelligent drill rig construction status codes

[0100] Code Construction status Code Construction status Code Construction status 0 Initialization 5 Slow drilling 10 Adding drill pipes 1 Resetting the gripper 6 Turning off water and stopping rotation 11 Tightening the rear thread 2 Rotating the power head and waiting for water 7 Clamping the front gripper 12 Opening the rear gripper 3 Drill cuttings sampling 8 Retracting the power head 13 Tightening the front thread 4 Fast drilling 9 Quick retraction of the power head 14 Opening the front gripper

[0101] Obtain the drill rig operating state in real time through sensors. When the drill rig enters the "Slow drilling" state (code 5), activate the CSTT system. The system starts to read the position of the drill rig power head relative to the fuselage. When the power head advances to half of the feeding stroke position, the drill rig control unit sends a start sampling command to the drill cuttings image processing unit.

[0102] After receiving the sampling instruction, the sampling process is as Figure 5As shown, the hydraulic system controls the drive cylinder of the sampling robotic arm to slowly retract, sending the sampling tray to the sampling position. During this period, the pressure sensor monitors the pressure of the drive cylinder. When the hydraulic pressure reaches 20 MPa, the system will consider that the drive cylinder is at the target position. Under the action of the flipping mechanism, the sampling tray flips to pour the chips into the weighing device. Subsequently, a continuous sampling time of 1 minute will be carried out. After sampling is completed, the hydraulic system controls the drive cylinder of the sampling robotic arm to slowly extend, moving the sampling tray to the image capture position. The condition for determining that the robotic arm has reached the image capture position is the same as that of the sampling position. During this process, the drill chips in the tray will be leveled. The drill chip sample will stay at the image capture position for 1 minute to drain the moisture. Subsequently, the image capture device starts to capture the drill chip image and transmits it to the drill chip image processing unit in the drill rig control. The drill chip image processing unit numbers the drill chips according to the sampling time and drilling depth information and stores them in the memory. (For details, reference can also be made to the patent "An Automatic Sampling, Weighing and Transportation System for Drill Chips")

[0103] This process innovatively realizes that the system automatically collects a drill chip sample every certain depth of drilling, achieving precise control of sampling in the spatial dimension; through a standardized sampling process, the sampling duration required for a single sampling is clarified, leaving time for the system to operate, and achieving precise control of sampling in the time dimension. The specific differences from the prior art are as follows:

[0104] (1) Standardization of the sampling timing is achieved

[0105] By analyzing the drilling construction process and the automatic drill chip sampling process, the entire process is divided into different drilling construction states. When the drill rig is in the "drilling" state, the automatic drill chip sampling process will be activated. At the same time, a displacement sensor located on the feed body of the drill rig detects the position of the rotary table. When the rotary table reaches the middle position of the feed stroke, the sampling command will be triggered. That is to say, for each drill rod drilled, when half of the drill rod is driven into the hole, the CSTT system will collect a drill chip image once. Thus, a uniform distribution of drill chip image samples along the drilling trajectory direction is achieved.

[0106] (2) Standardization of the sampling environment is achieved

[0107] To ensure the consistency of the imaging environment, the CSTT system integrates an image acquisition device, such as Figure 5As shown in the figure, it consists of an intrinsically safe camera, an independent light source assembly and a sampling tray. The image acquisition device creates a relatively closed environment between the intrinsically safe camera and the sampling tray, and provides independent lighting to eliminate the influence of the surrounding environment on image acquisition. In addition, the intrinsically safe camera uses the same shooting parameters for each shooting, such as shooting environment, shooting angle, camera settings and exposure time, which further ensures the consistency of the drill cuttings image sampling environment. The influence of the harsh environment in the coal mine, such as water mist, dust and darkness, on the quality of the drill cuttings image is eliminated.

[0108] (3) Standardization of the automatic sampling process of drill cuttings images

[0109] By standardizing the drilling cuttings image acquisition process, the time required for the sampling process is clarified, providing a basis for accurate sampling time control. At the same time, according to the accurate sampling time, the exact position of the power head on the drilling rig body during sampling can be determined, so as to calculate the position of the drill bit in the hole and achieve accurate positioning of the sampling point in space. The sampling process of the sampling point CSTT system is as follows Figure 5 shown.

[0110] (4) Prevents mixing of drill cuttings from different formations

[0111] In order to minimize the influence of mixed drill cuttings from different strata on the recognition accuracy during the drill cuttings image acquisition process, the present invention proposes the following method:

[0112] (1) After each drill pipe is completely drilled into the hole, the drilling rig keeps rotating for a period of time, and the wellbore is thoroughly flushed with drilling fluid to remove the remaining drill cuttings in the wellbore.

[0113] (2) To avoid the mixing of cuttings from different strata due to the on / off flow of flushing fluid during the loading and unloading of drill pipes, the cuttings sampling time is selected when each drill pipe has been drilled half its length.

[0114] (3) After the cuttings image sampling in the sampling plate is completed, the CSTT system will use the flipping device to flip the sampling plate, pour out the cuttings in the plate, and then proceed to the next sampling operation. At this time, even if there is a small amount of residual cuttings in the sampling plate, it will be covered by the newly flowing cuttings and will not be captured by the image acquisition device.

[0115] 1.4 Drilling retest

[0116] After the drilling construction is completed, the borehole is re-measured using a visual measuring instrument for mining boreholes to obtain accurate information on the borehole trajectory and the lithology changes along the borehole trajectory. Combined with the borehole logging data and the known geological information of the mine, the real lithology data at each sampling point is formed.

[0117] 1.5 Drill cuttings image preprocessing

[0118] In order to improve the ability of machine learning models to extract features of drill cuttings images and realize automatic acquisition of the proportions of different lithology samples in drill cuttings images, the present invention proposes a preprocessing method for broken coal and rock particle images, which includes a "preprocessing method for model training data sets" and a "preprocessing method for normal construction data sets".

[0119] The "preprocessing method of model training data set" will be used in this stage; the "preprocessing method of normal construction data set" will be used in the model deployment and implementation stage.

[0120] Preprocessing method of model training data set:

[0121] (1) Screening of original images

[0122] Based on the borehole retest data, the drill cuttings image samples collected from the interface of two different lithology formations are eliminated. This step is taken to reduce the impact of drill cuttings mixing on the accuracy of lithology labeling of training samples.

[0123] (2) Establishment of model training data set:

[0124] Initial cropping: The original drill cuttings image was cropped to remove invalid information such as the edge of the sampling container.

[0125] Secondary cropping: The original image is cropped secondary according to the input image requirements of the selected image recognition deep learning model, such as 224 pixels × 224 pixels.

[0126] Numbering: Number the drill cuttings image blocks obtained after secondary cropping, such as Figure 7 (a) shown.

[0127] Lithology labeling: Label the lithology of each image based on the real experimental data of the drill cuttings image collection points obtained previously, such as Figure 7 (b) Then the image blocks are divided into different groups according to different lithologies to form the model training data set.

[0128] (3) Establishment of model test data set:

[0129] 80% of the data in the randomly rotated model training data set is used for training the machine learning model, and 20% of the data forms the model test data set, which is used to test the accuracy changes during the machine learning model training process.

[0130] 1.6 Image enhancement and establishment of training data set

[0131] Using runtime enhancement technology, when reading the original training images, a certain transformation or combination of transformations shown in Table 2 is performed on them with a given probability of 80%. Then the transformed images are added to the current training batch for model training. After training, the transformed images are not saved.

[0132] Table 2 Image Enhancement Parameters

[0133]

[0134] 1.7 Training and Optimization of Individual Classification Models

[0135] Common deep learning pre-trained models for image recognition include classic convolutional networks (such as ResNet, DenseNet), lightweight models (such as MobileNet, EfficientNet), and attention mechanism-based models (such as Vision Transformer and Swin Transformer). In this stage, the model training dataset and the model test dataset are used to test these common models and select deep learning pre-trained models with high recognition accuracy for drill cuttings images. Then, using transfer learning, these pre-trained models are intensively trained. By selecting different optimization algorithms, changing the model structure, adjusting the initial learning rate and validation frequency, etc., the recognition accuracy of the individual classification models is optimized, and a model with both recognition accuracy and speed is selected.

[0136] 1.8 Construction of Ensemble Learning Model

[0137] Based on the soft voting ensemble learning method, multiple individual classification models that have been intensively trained before are used to establish an ensemble learning model for lithology recognition based on drill cuttings images. By adjusting the weights of each model, the corresponding relationship between the weight sum and the recognition accuracy of the ensemble model is formed, so as to obtain the optimal weight combination and construct the optimal ensemble learning model. The model test dataset is used to test, optimize and evaluate the model to ensure that the drill cuttings image samples can be accurately classified. After passing the test, the training of the lithology recognition model is completed.

[0138] The model deployment and implementation stage specifically includes the following content:

[0139] 2.1 Integration of Ensemble Learning Model

[0140] In the deployment and implementation stage, the model is integrated into the control system of the intelligent drill rig. Drill cuttings image data is read from the drill rig controller in real time and recognized. The recognition results are then transmitted and stored in the storage unit of the intelligent drill rig controller, and are also displayed on the screen of the drill rig's control system, such as the screen of the remote controller.

[0141] 2.2 Automatic Acquisition of Drill Cuttings Images

[0142] During the drilling operation, the intelligent drill will work in coordination with the CSTT system. For each drill pipe installed, a cuttings image is collected and transmitted to the drill controller.

[0143] 2.3 Image preprocessing

[0144] Preprocessing method for normal construction data:

[0145] After the model is trained, optimized and integrated into the drill system, the preprocessing method for the original cuttings image is slightly different from that in the model training stage. To enable the recognition model to comprehensively extract features from the cuttings image, the collected image data is first cropped as Figure 8 shown. The cropping requirements are the same as those described above. Then, each image patch is randomly rotated to generate multiple new image patches. These newly generated image patches, together with the original image patches, constitute the lithology recognition dataset for this specific image patch.

[0146] 2.4 Lithology recognition and result output

[0147] After the recognition of a single cuttings image is completed, the number and recognition accuracy of the cuttings image patches identified as different lithologies are counted. By selecting the category with the highest weighted average probability in the lithology recognition dataset, the lithology recognition result of each image patch is determined. According to the proportion of the image patches with the same lithology type in the total number of image patches, the distribution proportion of this lithology in the cuttings is determined.

[0148] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solutions of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0149] In addition, it should be noted that, in the case of no contradiction, the various specific technical features described in the above specific embodiments can be combined in any suitable manner. To avoid unnecessary repetition, the present invention will not separately describe various possible combination methods.

[0150] Furthermore, any combination can be made between various different embodiments of the present invention, as long as it does not violate the idea of the present invention, and it should also be regarded as the content disclosed by the present invention.

Claims

1. An intelligent identification method for formation lithology of rotary drilling in coal mines, characterized in that, The method includes: Step 1, model training stage: trajectory design of the demonstration borehole, joint commissioning of the system, borehole construction and automatic collection of drill cuttings images, borehole remeasurement, preprocessing of drill cuttings images, image enhancement and establishment of the training dataset, training and optimization of the individual classification model, construction of the ensemble learning model; Step 2, deployment and implementation stage: integration of the ensemble learning model, automatic collection of drill cuttings images, image preprocessing, lithology identification and result output.

2. The intelligent identification method for formation lithology of underground rotary drilling in coal mines according to claim 1, wherein The trajectory design of the demonstration borehole covers the typical strata in the mining area and potential geological structures to be identified in the future to collect sufficient data for training the lithology identification machine learning model.

3. The intelligent identification method for formation lithology of underground rotary drilling in coal mines according to claim 1, characterized in that, In the joint commissioning of the system, the system includes an intelligent drill rig, a CSTT system, and a mine borehole visualization measurement instrument; The intelligent drill rig includes: a drilling system, an angle adjustment system, a control system, a crawler walking system, an automatic drill pipe loading and unloading system, a pump station, and a controller; The CSTT system includes a hole mouth device and a main unit. The main unit includes a cover plate assembly, an image acquisition device, a weighing device, a drill cuttings conveying device, a frame assembly, and a slag discharge pipe; the cover plate assembly, the image acquisition device, the weighing device, and the drill cuttings conveying device are arranged in the frame assembly from top to bottom; the cover plate assembly includes a cover plate and a slag inlet on it; the hole mouth device is arranged at the borehole mouth, one end of which is connected to the borehole mouth, and the other end is connected to the slag inlet on the cover plate assembly through the slag discharge pipe; the image acquisition device includes an image acquisition device, a sampling tray, and a sampling robotic arm, and the sampling robotic arm can drive the sampling tray to move between the drill cuttings collection point below the slag inlet and the image acquisition point below the image acquisition device; The mine borehole visualization measurement instrument can efficiently obtain high-definition videos, three-dimensional trajectories, and coal and rock properties of underground boreholes to quickly identify the lithology of the strata in the roadway trajectory direction; The joint commissioning of the system includes: installing the intelligent drill rig and the CSTT system in the drill yard, completing the pipeline connection and conducting joint commissioning; adjusting the drill rig to the designed inclination angle, azimuth angle, and opening position, and adjusting the drill rig body to be stable; installing an opening drill bit on the drill rig, completing the borehole opening and installing the outer hole mouth pipe; installing the hole mouth device of the CSTT system at the borehole mouth, inserting the hole mouth pipe of the hole mouth device into the outer hole mouth pipe of the borehole, and sealing the annulus; connecting the hole mouth device and the main unit of the CSTT system with the slag discharge pipe; connecting each actuator of the CSTT system to the control valve with the corresponding function on the drill rig through a hydraulic hose; debugging to ensure that the drill rig and the CSTT system operate and data transmission are normal.

4. The intelligent recognition method for formation lithology of downhole rotary drilling in coal mines according to claim 1, characterized in that, The borehole construction and automatic collection of drill cuttings images include: When sorting out the automatic drilling construction process of the intelligent drilling rig and the drill cuttings image acquisition process of the CSTT system, the entire drilling construction process is divided into 15 drilling rig operation states; the operation state of the drilling rig is obtained in real time through sensors, the system starts to read the position of the power head of the drilling rig relative to the fuselage, and the drilling rig control unit sends a start sampling command to the drill cuttings image processing unit; the hydraulic system controls the driving cylinder of the sampling manipulator to slowly retract, and sends the sampling tray to the sampling position; during this period, the pressure sensor monitors the pressure of the driving cylinder, and when the hydraulic pressure reaches 20 MPa, the system considers that the driving cylinder is in the target position; the sampling tray flips to pour the cuttings into the weighing device; after sampling, control the sampling manipulator to move the sampling tray to the image capture position; during this process, the drill cuttings in the sampling tray are leveled; the drill cuttings sample stays at the image capture position to drain the moisture; subsequently, the image collector starts to capture the drill cuttings image and transmits it to the drill cuttings image processing unit in the drilling rig control.

5. The intelligent identification method for formation lithology of underground rotary drilling in coal mines according to claim 1, characterized in that, The borehole remeasurement includes: after the borehole construction is completed, use a mine borehole visualization measurement instrument to remeasure the borehole to obtain accurate borehole trajectory information and formation lithology change information along the borehole trajectory direction; combine the logging data of the borehole and the known geological information of the mine to form the true lithology data at each sampling point. The preprocessing of the drill cuttings image includes: screening of the original image, establishment of the model training data set, and establishment of the model test data set. The screening of the original image: According to the borehole remeasurement data, eliminate the drill cuttings image samples collected from the interface between two different lithology formations. The establishment of the model training data set includes: Primary cropping: Crop the original drill cuttings image to remove invalid information such as the edge of the sampling container. Secondary cropping: According to the requirements of the selected image recognition deep learning model for the input image. Numbering: Number the drill cuttings image blocks obtained after secondary cropping. Lithology annotation: Annotate the lithology of each image according to the true experimental data of the drill cuttings image acquisition point obtained previously, and then divide the image blocks into different groups according to different lithologies to form the model training data set. The establishment of the model test data set includes: randomly select 80% of the data in the model training data set for the training of the machine learning model, and 20% of the data form the model test data set for testing the accuracy change during the training process of the machine learning model.

6. The intelligent identification method for formation lithology of downhole rotary drilling in coal mines according to claim 1, characterized in that The establishment of the image enhancement and training data set includes: Adopt the runtime enhancement technology, when reading the original training image, transform or combine transform it with a given probability of 80%; then add the transformed image to the current training batch for model training; after the training is over, the transformed image will not be saved. The training and optimization of the individual classification model: Test these common models using the model training dataset and the model testing dataset, and select a deep learning pre-trained model with high recognition accuracy for cuttings images. Then, use transfer learning to perform reinforcement training on these pre-trained models; optimize the recognition accuracy of individual classification models by selecting different optimization algorithms, changing the model structure, adjusting the initial learning rate and validation frequency, etc., and select a model with both recognition accuracy and speed.

7. The intelligent identification method for formation lithology of underground coal mine rotary drilling according to claim 1, characterized in that Construction of the ensemble learning model: Based on the soft voting ensemble learning method, use multiple individual classification models that have been well-trained in the previous reinforcement training to establish an ensemble learning model for lithology recognition based on cuttings images; by adjusting the weights of each model, form the corresponding relationship between the weight sum and the recognition accuracy of the ensemble model, so as to obtain the optimal weight combination and construct the optimal ensemble learning model; use the model testing dataset to test, optimize and evaluate the model to ensure accurate classification of cuttings image samples; after passing the test, complete the training work of the lithology recognition model.

8. The intelligent identification method for formation lithology of underground rotary drilling in coal mines according to claim 1, characterized in that Ensemble of the ensemble learning model: In the deployment and implementation stage, the model is integrated into the control system of the intelligent drilling rig; Read the cuttings image data from the drilling rig controller in real time and perform recognition; the recognition results will be transmitted and stored in the storage unit of the intelligent drilling rig controller, and at the same time displayed on the screen of the control system of the drilling rig; Automatic acquisition of the cuttings images: During drilling construction, the intelligent drilling rig and the CSTT system work in coordination. For each drill pipe constructed, a cuttings image is collected and transmitted to the drilling rig controller.

9. The intelligent identification method for formation lithology of underground rotary drilling in coal mines according to claim 1, wherein, Image preprocessing: Preprocessing method for normal construction data: After training, optimizing the model and integrating it into the drilling rig system, the preprocessing method of the original cuttings images is slightly different from that in the model training stage; first, the collected image data is cropped, and then each image patch is randomly rotated to generate multiple new image patches; these newly generated image patches and the original image patches together constitute the lithology recognition dataset for this specific image patch.

10. The intelligent identification method for formation lithology of downhole rotary drilling in coal mines according to claim 1, characterized in that, Lithology recognition and result output: After the recognition of a single cuttings image is completed, count the number and recognition accuracy of the cuttings image patches recognized as different lithologies, and determine the lithology recognition result of each image patch by selecting the category with the highest weighted average probability in the lithology recognition dataset; determine the distribution ratio of this lithology in the cuttings according to the proportion of the image patches with the same lithology type in the total number of image patches.

Images