A Zoned Cutting Control Method for Roadheaders Based on Coal and Rock Identification

By using dual-motor frequency conversion control and multi-source data acquisition technology, the roadheader has achieved adaptive cutting and anchoring of coal and rock zones, solving the problems of high energy consumption and low identification accuracy in underground hard rock tunneling, improving equipment stability and lifespan, and optimizing working condition monitoring.

CN123109690APending Publication Date: 2026-11-17YULINYUSHENMEITANYUSHUWAN COAL MINE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611396543.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-09
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing roadheader and anchor machine suffers from high energy consumption, large vibration, and short equipment life in underground hard rock tunneling. It also has low coal and rock identification accuracy, fixed anchoring parameters leading to unstable surrounding rock, and lacks real-time monitoring and fault analysis capabilities.

Method used

The system employs a dual-motor variable frequency control system, combined with multi-source data acquisition and deep learning algorithms, to identify coal and rock zones in real time, adaptively adjust cutting and anchoring parameters, enhance the machine body structure, and achieve data visualization and storage.

Benefits of technology

It improves the accuracy of coal and rock identification, reduces energy consumption and mechanical wear, extends equipment life, optimizes anchoring effect, and provides full-process operating condition monitoring and fault analysis capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123109690A_ABST
    Figure CN123109690A_ABST
Patent Text Reader

Abstract

This invention provides a zoned cutting control method for roadheader / anchor machine based on coal and rock identification, belonging to the technical field of zoned cutting control for roadheader / anchor machine. To overcome the shortcomings of low coal and rock identification accuracy, lack of hardness quantification, insufficient machine structure and power performance, and poor anchoring adaptability during roadheader / anchor machine operation, an auxiliary cutting motor is added to the existing main cutting motor of the roadheader / anchor machine and debugged. Images of the coal and rock cutting surface in the roadway are acquired by explosion-proof industrial cameras deployed underground. Vibration signals, operating current signals, and anchoring borehole pressure signals of the two motors are simultaneously acquired by sensors deployed underground. After filtering and noise reduction preprocessing by the lower-level computer, the data are uploaded to the upper-level computer. Multimodal feature extraction of the cutting surface images is performed using a deep feature multiplexing network, and the vibration and operating current signals of the motors are obtained using a multi-scale one-dimensional residual network. The upper-level computer then fuses the data. This invention is applied to the zoned cutting control of roadheader / anchor machines.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention provides a method for zoned cutting control of roadheader based on coal and rock identification, belonging to the field of zoned cutting control technology for roadheader. Background Technology

[0002] As a key piece of equipment for simultaneous tunneling and anchoring in underground mines, the integrated tunneling and anchoring machine's intelligent control capabilities, adaptability to geological conditions, and structural strength directly affect the quality of tunnel formation and the safety of underground mining operations. However, several technical challenges remain for concurrent tunneling and anchoring operations underground, significantly limiting the application of intelligent tunneling equipment in coal mines. These challenges include:

[0003] Underground hard rock cutting conditions, rock breaking energy consumption is high and tunneling impact load is large. The original cutting power of conventional roadheader is insufficient to meet the needs of hard rock tunneling. The vibration amplitude of the whole machine increases significantly during rock cutting, which accelerates the fatigue wear of the machine body structure and shortens the overall service life of the equipment. At the same time, long-term high vibration conditions will reduce the operational stability of the on-board hydraulic components and electrical control components, ultimately resulting in frequent equipment failures and a decrease in the effective uptime rate underground. Therefore, coal and rock identification in the working area is very important during roadheader construction. However, the traditional machine vision algorithms currently used for identification will have large deviations in identifying the coal and rock interface under the interference of high dust concentration, uneven light distribution, rough rock surface, and heavy water mist underground, and cannot accurately detect the actual hardness of coal and rock.

[0004] Currently, the anchoring support system of the tunneling and anchoring machine adopts fixed operating parameters and does not adjust the process parameters automatically according to the change of the cutting surface hardness. For soft coal seams and soft rock layers, excessive anchoring pressure and torque will directly squeeze and damage the surrounding rock of the borehole, causing the surrounding rock to crack and the borehole wall to collapse. For hard rock layers, the original anchoring parameters are insufficient, the anchoring fit of the anchor bolts is not good, and the support anchoring force is not up to standard, which poses a safety hazard to the stability of the surrounding rock in the later stage of the roadway.

[0005] In addition, the existing tunneling and anchoring machines do not have a supporting operation monitoring platform. On-site personnel cannot intuitively view core operating information such as cutting motor vibration data and changes in operating current, which is not conducive to subsequent equipment fault repair, tunneling condition review, and construction process optimization. Summary of the Invention

[0006] To address the technical problems existing in the background art, the present invention provides a method for zoned cutting control of roadheaders based on coal and rock identification, comprising the following control steps:

[0007] Step S1: On the basis of the original main cutting motor of the roadheader, an auxiliary cutting motor is set up, and the dual-motor frequency conversion control system is installed and debugged; the main beam of the roadheader body, the cutting arm base, and the anchoring frame are reinforced by welding, strengthening with reinforcing ribs, and damping shock absorption components are installed.

[0008] Step S2: During underground tunneling and anchoring operations, images of the coal and rock cutting face in the roadway are collected by explosion-proof industrial cameras deployed underground. Vibration signals, working current signals, and anchoring borehole pressure signals of the dual motors are collected synchronously by vibration sensors, current transformers, and pressure sensors deployed underground. After filtering and noise reduction preprocessing by the lower computer, the signals are uploaded to the upper computer.

[0009] Step S3: Multimodal feature extraction is performed on the cut surface image through a deep feature reuse network. The vibration signal and operating current signal of the motor are obtained using a multi-scale one-dimensional residual network. The data are then fused by the host computer to generate fused features.

[0010] Step S4: Input the fused features generated in step S3 into the Gaussian process regression model, and combine them with the calibration dataset to output the hardness value and confidence level of each coal and rock partition in real time.

[0011] Step S5: Establish the judgment rules for pure coal area, soft rock area, hard rock area and coal-rock mixed area as an expert rule base, and then divide the cutting surface into pure coal area, soft rock area, hard rock area and coal-rock mixed area according to the output of step S3 and step S4.

[0012] Step S6: The host computer displays the working current waveform of the dual motors, vibration signal, anchoring drilling pressure, current zone and hardness, and current excavation and anchoring parameters in real time. Each data is bound with a timestamp and stored in the local database and the cloud.

[0013] Step S7: Based on the coal and rock zoning results obtained in step S5, adaptively adjust the output power, rotation speed, feed speed, and single cutting depth of the dual motors to achieve differentiated cutting control for different geological zones;

[0014] Step S8: Based on the real-time detection results of coal and rock zoning, adaptively adjust the drilling speed, drilling pressure, anchoring depth, and anchor bolt torque of the anchoring operation to achieve differentiated anchoring control for different geological zones.

[0015] The specific method for step S1 is as follows:

[0016] The auxiliary cutting motor and the main cutting motor are symmetrically arranged on both sides of the cutting drum. The main cutting motor is used for the main cutting operation of coal and rock, and the auxiliary cutting motor is used for auxiliary cutting in the rocky area. The two motors are equipped with independent frequency conversion control modules, and the output parameters are adjusted independently according to the coal and rock zones.

[0017] The specific method for step S2 is as follows:

[0018] The cut surface image is specifically captured by an explosion-proof industrial camera mounted on the cutting arm, acquiring real-time visible light image data of the cut surface, expressed as:

[0019] ;

[0020] In the formula, This is a single-frame cut surface image data; The real number field indicates that each element in the matrix is ​​a real number; The height of the image. 3 represents the width of the image, and 3 represents the number of channels in the image, corresponding to the red, green, and blue color channels of a color image.

[0021] The vibration signal is specifically acquired using a triaxial accelerometer to collect the vibration acceleration during the cutting of the motor housing, with a sampling frequency of... This constitutes a three-channel timing signal. The expression is:

[0022] ;

[0023] ;

[0024] In the formula, For time The vibration signal matrix is ​​the independent variable; 3 represents the real number field; 3 represents the number of channels, corresponding to the vibration components in the three mutually perpendicular directions of X, Y, and Z acquired by the triaxial accelerometer. This represents the number of sampling points contained in the vibration signal segment. The sampling frequency of the vibration signal; This refers to the duration of the segment;

[0025] The operating current signal is specifically acquired through a Hall current sensor. By collecting the effective values ​​of the three-phase current of the cutting motor, a single-channel timing signal is formed. The expression is:

[0026] ;

[0027] ;

[0028] In the formula, For time The current signal vector is the independent variable, and a time segment synchronized with the vibration signal is extracted. For the real number field; This represents the number of sampling points contained in the current signal segment. The sampling frequency of the current signal; This refers to the duration of the segment.

[0029] The specific method for step S3 is as follows:

[0030] Multimodal feature extraction of the cut surface image is performed using the DenseNet network, including:

[0031] Construct a densely connected convolutional network. The basic unit of the convolutional network is composed of several feature reuse modules stacked together. The output of the feature multiplexing module Defined as:

[0032] ;

[0033] In the formula, This is a hierarchical index for the feature reuse module. =1,2,…, , This represents the total number of layers in the module. The initial input feature map for the network is the cut surface image. The feature tensor obtained after initial convolution; For the first The feature map tensor output by the layer feature reuse module is the feature map tensor of that layer after passing through a composite function. The new feature representation generated after the transformation; For the first The feature map output from the -1 layer feature reuse module is concatenated with the outputs of earlier layers and fed into the ... layer;

[0034] The feature vector of the output image of the convolutional network The expression is:

[0035] ;

[0036] In the formula, For the process The feature map output by the last layer of the network after the feature reuse module; The dimension of the image feature vector;

[0037] The vibration signal and operating current signal of the motor are obtained using a multi-scale one-dimensional residual network, including:

[0038] Vibration signals and current signal Robust features are extracted using multi-scale one-dimensional convolutional modules, where:

[0039] Vibration signal First, the data undergoes short-time energy normalization and low-pass filtering. Then, it is input into three layers of residual convolutional blocks, each containing parallel 1×3, 1×5, and 1×7 convolutional kernels. Finally, multi-scale information is fused through gated linear units to output a vibration feature vector. ,in The dimension of the vibration feature vector is determined by the number of output channels of the last convolutional layer;

[0040] Current signal After processing, the current characteristic vector is obtained. ,in The dimension of the current feature vector is used to preserve the statistical properties of the amplitude domain, while also calculating handcrafted features. :

[0041] ;

[0042] In the formula, The kurtosis of the vibration signal; The root mean square of the vibration signal; The fluctuation rate of the current signal; The number of handcrafted features; Indicates transpose;

[0043] The various data will be merged, including:

[0044] The above features are concatenated to form a comprehensive feature vector. :

[0045] ;

[0046] In the formula, The total dimension of the fused features numerically satisfies = .

[0047] The specific method for step S4 is as follows:

[0048] A Gaussian process regression model was used, with the fused features as input. The output is the coal and rock hardness value. This yields a numerical prediction of actual hardness with the ability to express uncertainty, including:

[0049] Modeling a Gaussian process:

[0050] The hardness value follows a Gaussian prior process, expressed as:

[0051] ;

[0052] In the formula, For different fusion characteristics; For fusion features The corresponding hardness value;

[0053] Zero-mean function The kernel function is a composite kernel capable of capturing the nonlinear effects of different modes. Including radial base core polynomial kernel Periodic nuclei The expression is:

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] In the formula, For different fusion characteristics; for Transpose of; This serves as the variance scale for the radial basis kernel, controlling the amplitude of this kernel component; is the variance scale of the polynomial kernel; The variance scale for the periodic kernel; The periodic parameter represents the characteristic period of the load fluctuation of the cutting drum; The length scale parameter of the radial basis kernel determines the smoothness of the function and the rate at which the correlation in the input space decays; The length scale parameter of the periodic kernel affects the smoothness of the periodic components; For the constant bias term of the polynomial kernel, This controls the weights of lower-order interactions; The order of the polynomial kernel is a positive integer, which determines the complexity of the nonlinear interaction.

[0059] Training and prediction include:

[0060] The collected calibration dataset Input, where For the first The fusion features of each training sample The logarithmic marginal likelihood is the true hardness measured in the laboratory using on-site sampling:

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] Iterative adjustment via gradient optimizer and Until convergence;

[0066] For real-time new samples The predicted average hardness value and variance for:

[0067] ;

[0068] ;

[0069] This yields the predicted hardness value. and forecast uncertainty ;

[0070] In the formula, This is a column vector composed of training hardness values; The total number of training samples; For hyperparameter set; The input matrix; The matrix of the composite kernel for training input, i.e. ,for Symmetric positive definite matrix; Let V be the variance of the Gaussian observation noise.

[0071] The specific method for step S5 is as follows:

[0072] The input variables for defining an expert rule base include:

[0073] Hardness prediction mean ;

[0074] Hardness prediction standard deviation , indicating uncertainty;

[0075] Visual proportion of coal and rock The category activation map output by the dense network calculates the proportion of coal-bearing regions in the image. ;

[0076] Vibration Amplitude Index , indicating the cutting impact strength;

[0077] Current fluctuation index ;

[0078] The output of the expert rule base is the partitioning results and confidence scores. The zoning results include: pure coal zone, soft rock zone, hard rock zone, and mixed coal-rock zone. The zoning method is as follows:

[0079] Define linguistic values ​​and membership functions for each variable:

[0080] The language values ​​are low, low-medium, medium, high-medium, and high.

[0081] The language value is small, medium, or large;

[0082] The language values ​​are: pure coal, coal with little rock, rock with little coal, hard rock;

[0083] The language value is weak, medium, or strong;

[0084] The language value is stable, with moderate fluctuations, or with drastic fluctuations.

[0085] Define the expert rule base rules as follows:

[0086] Rule 1: If For low, For pure coal, For the weak, For stability, it is a pure coal zone;

[0087] Rule 2: If For medium and low It is characterized by abundant coal and scarce rock. For China If the fluctuation is moderate, it indicates a coal-rock mixture zone;

[0088] Rule 3: If For the middle and If it is large, then further observation is needed. ,like If there is more rock than coal, it is a coal-rock mixed zone. If the tendency is towards hard rock, then it leans towards soft rock areas;

[0089] Rule 4: If For middle and high school It is characterized by abundant rocks and scarce coal. For strength, If the fluctuations are violent, it indicates a soft rock area;

[0090] Rule 5: If For high, For hard rock, For strength, If the fluctuations are violent, it indicates a hard rock zone;

[0091] Rule 6: When When the value is large, the overall confidence level of partition determination decreases, and it relies more on... and The strong evidence rule shall be used to make the decision;

[0092] Performing fuzzy reasoning and defuzzification:

[0093] The Mamdani minimum inference method and the centroid method are used to defuzzify the data, outputting continuous partition membership scores. Finally, based on the maximum membership principle, the partition categories are determined: pure coal zone, soft rock zone, coal-rock mixed zone, and hard rock zone, and the confidence score of each category is also output. .

[0094] The specific method for step S7 is as follows:

[0095] Based on the cutting surface partitioning results output by the DenseNet network and the real-time coal and rock hardness values ​​obtained from Gaussian process regression, combined with the modified dual-motor cooperative power structure, corresponding tunneling power and operating parameters are used for partitioned cutting control, including:

[0096] For pure coal areas: the system only turns on the main cutting motor to operate independently, while the auxiliary cutting motor enters a sleep standby state, increasing the drum rotation speed and the overall machine feed speed, and reducing the single cutting depth;

[0097] For soft rock areas: the system controls two cutting motors to operate in coordination at low power, while adjusting the machine feed speed to prevent excessive feed from causing local collapse of the surrounding rock, avoid over-excavation of the cross section, and ensure the flatness of the roadway formation;

[0098] For hard rock areas: The system controls dual motors to output full power synchronously, reducing the overall machine feed speed and increasing the single cutting depth;

[0099] For coal-rock mixed zones: different cutting parameters are used for different hardness levels.

[0100] The specific method for step S8 is as follows:

[0101] An adaptive anchoring control scheme is adopted, dividing the cutting surface into high-hardness rock zone, low-hardness coal zone, low-hardness soft rock zone, and medium-hardness coal-rock mixed zone. The following anchoring operation parameters are used for control of different zones:

[0102] For high-hardness hard rock areas: increase the anchoring drilling speed, increase the drilling feed pressure, and simultaneously increase the anchoring depth and anchor bolt locking torque;

[0103] For low-hardness coal areas and soft rock areas: reduce drilling speed and feed pressure, reduce anchor bolt locking torque, and shorten anchoring depth;

[0104] For coal-rock mixed zones: instead of using uniform anchoring parameters, corresponding anchoring parameters are assigned according to the local real-time hardness of the cut surface.

[0105] The present invention has the following advantages over the prior art:

[0106] I. Optimize the existing coal and rock identification algorithm architecture to improve the identification accuracy of complex underground working conditions: This invention adopts an algorithm that combines a lightweight DenseNet feature extraction network, a Gaussian process, and an expert rule base, and integrates three types of multi-source operating data: on-site images, equipment vibration, and motor current, to predict the hardness of the cutting surface and divide the coal and rock into zones.

[0107] II. Improve the overall power and load-bearing structure to reduce mechanical wear: This invention addresses the problems of insufficient power, significant deformation of the machine body under heavy load, and severe vibration in traditional single-motor tunneling and anchoring machines for hard rock tunneling. It adds an auxiliary cutting motor and strengthens and optimizes the structure and vibration reduction of key stress-bearing parts such as the main beam of the machine body and the cutting base.

[0108] 3. Full-process operating condition monitoring and data storage to achieve traceability of equipment operation and maintenance: The control system provided by this invention can display the vibration waveform and working current waveform of the dual motors in real time, making the internal operating status of the equipment visible; at the same time, all operating data, coal and rock identification data, and on-site operation parameters are backed up locally and in the cloud, which facilitates the understanding of the working status of the roadheader and also facilitates the optimization of the cutting control method and control parameters of the roadheader.

[0109] IV. Achieving adaptive control of cutting and anchoring dual system parameters: The control system provided by this invention divides the cutting surface into pure coal zone, soft rock zone, hard rock zone and coal-rock mixed zone. Different zones correspond to different cutting and anchoring parameters. If it is in the coal-rock mixed zone, the hardness of different areas in the entire cutting surface is also different. At this time, different cutting and anchoring parameters need to be used according to the hardness of the current area. The switching of cutting and anchoring parameters should be smooth.

[0110] V. Energy saving and consumption reduction, low operation and maintenance costs: This invention can realize dynamic adaptive power ratio and operating parameter control, effectively reduce equipment ineffective energy consumption and heavy load impact loss, reduce equipment mechanical wear by shock absorption and reinforcement structure, extend equipment service life, and reduce downhole equipment operation and maintenance costs and downtime losses. Attached Figure Description

[0111] The present invention will be further described below with reference to the accompanying drawings:

[0112] Figure 1 This is a flowchart illustrating the steps of the roadheader zone cutting control method based on coal and rock identification according to the present invention.

[0113] Figure 2 This is a schematic diagram of the structure of the tunneling and anchoring machine zone cutting control system based on coal and rock identification according to the present invention. Detailed Implementation

[0114] To enable the use of different cutting and anchoring parameters in different tunneling and anchoring operation areas, this invention divides the cutting surface of the tunneling and anchoring machine's operation area into pure coal zones, soft rock zones, hard rock zones, or coal-rock mixed zones based on an improved DenseNet network, Gaussian process, and expert rule base. Using different cutting and anchoring parameters for different zones can reduce vibrations during tunneling and anchoring and minimize unnecessary power consumption. This invention also reinforces the structure of the tunneling and anchoring machine to reduce mechanical wear. To enable visualization of the machine's operating status and traceability of faults, the host computer displays data such as cutting motor current, machine vibration waveform, cutting surface hardness value, and drilling pressure in real time. All data and operation parameters are simultaneously backed up locally and in the cloud.

[0115] like Figure 1 As shown in the figure, this invention provides a method for zoned cutting control of roadheader based on coal and rock identification, which mainly includes the following steps:

[0116] Step S1: Structural optimization and modification of the roadheader. Rock cutting consumes a large amount of energy and is subject to strong impact loads. In order to improve the cutting capacity and service life of the roadheader, hardware modifications were made to the roadheader. Auxiliary cutting motors were symmetrically added to the original main cutting motor, and the dual-motor frequency conversion control system was installed and debugged. The main beam of the machine body, the cutting base, and the anchoring frame were reinforced by welding, strengthening with reinforcing ribs, and damping and shock absorption components were installed.

[0117] Step S2: Real-time acquisition and synchronization of multi-source working condition data. During underground tunneling and anchoring operations, high-definition explosion-proof industrial cameras are used to acquire real-time images of the coal and rock cross-sections in the roadway. Vibration sensors, current transformers, and pressure sensors are used to synchronously acquire vibration signals from dual motors, working current signals, and anchoring borehole pressure signals. After filtering and preprocessing by the lower-level computer, the signals are uploaded to the upper-level computer in real time. The three types of signals from the vibration sensor, current transformer, and pressure sensor are aligned and segmented on the time axis based on the rotation cycle of the cutting drum, forming "cutting surface" and "signal segment" sample pairs.

[0118] Step S3: Multimodal feature extraction. Image features are extracted using a deep feature reuse network (an improved DenseNet structure), motor vibration and current signals are obtained using a multi-scale one-dimensional residual network, and finally all data are fused to generate fused features;

[0119] Step S4: Numerical regression of hardness based on Gaussian process. The fused features from step S3 are input into the Gaussian process regression model, and combined with the calibration dataset, the hardness values ​​and confidence levels corresponding to each coal and rock zone are output in real time.

[0120] Step S5: Expert rule base for coal-rock zone division. Establish rules for determining pure coal zone, soft rock zone, hard rock zone, and mixed coal-rock zone. Based on the outputs of steps S3 and S4, divide the cutting surface into pure coal zone, soft rock zone, hard rock zone, and mixed coal-rock zone.

[0121] Step S6: Real-time visualization of waveforms and storage of historical data on the host computer. To enable operators to have an intuitive understanding of the working status of the roadheader, the host computer will display real-time data such as motor current waveform, motor vibration wave, drilling pressure, current zone and hardness, and current roadheader parameters. This data will also be stored in a local database and in the cloud for easy access to subsequent working condition queries, playback of operating data, and fault tracing analysis.

[0122] Step S7: Adaptive Cutting Control for Coal and Rock Zones. The system adjusts the cutting parameters based on the zone results from Step S5: If the zone result is a pure coal zone, due to its low hardness, only the main motor is needed; continuing to operate the auxiliary motor would result in unnecessary energy consumption. If the zone result is a soft rock zone, due to its relatively hardness, using only the main motor would not provide sufficient cutting force; the auxiliary motor needs to provide some cutting force, with both motors operating at low power and the feed speed appropriately reduced. If the zone result is a hard rock zone, due to its high hardness, both motors need to operate at full power, while the feed speed is reduced, but the cutting depth is increased. If the zone result is a mixed coal and rock zone, the hardness of different areas of the cutting surface varies. In this case, the corresponding cutting parameters need to be used according to the hardness of the current area. Different hardnesses correspond to different cutting parameters, and the switching between different cutting parameters should be smooth; otherwise, it will cause severe vibration of the entire machine, leading to rock wall fracture.

[0123] Step S8: Adaptive Anchoring Control for Coal and Rock Zoning. Just like cutting, the anchoring parameters also need to be automatically adjusted according to the zoning results and hardness, but they are slightly different from cutting. The specific method for adjusting the anchoring parameters is as follows: (1) If it is a hard rock zone, due to the high hardness, the drilling pressure, motor speed, anchoring depth and anchor cable pre-tightening torque need to be increased; (2) If it is a pure coal zone and a soft rock zone, due to the moderate hardness, the drilling pressure, motor speed, anchoring depth and anchor cable pre-tightening torque can be appropriately reduced; (3) If it is a coal and rock mixed zone, the hardness is changing. At this time, the drilling pressure, motor speed, anchoring depth and anchor cable pre-tightening torque need to be adjusted according to the hardness value of the current zone, but the adjustment process should be gradual.

[0124] like Figure 2As shown, to achieve the above-mentioned zoned cutting control method for the roadheader, this invention also improves and optimizes the corresponding structure of the roadheader. By installing a motor on each side of the cutting drum, with the main cutting motor already present, auxiliary cutting motors are symmetrically installed on top of it. In order to increase the cutting capacity on cutting surfaces with high hardness, this can reduce the mechanical wear of the main cutting motor. The aforementioned addition of auxiliary cutting motors brings a problem, that is, resonance will occur when the two motors work simultaneously. To address this, this invention adds damping and vibration reduction components at the motor hinge points. In addition, in order to reduce the vibration of the entire machine during operation and improve the service life of the entire machine, this invention also reinforces the main beam of the machine body and the stress concentration points of the frame, and uses an integral casting molding process to manufacture the cutting base.

[0125] Furthermore, to overcome the shortcomings of existing technologies such as low accuracy in coal and rock identification, lack of hardness quantification capability, fixed tunneling and anchoring parameters, untraceable equipment operating conditions, insufficient machine structure and power performance, and poor anchoring adaptability, this invention provides a zoned cutting control method for tunneling and anchoring machines based on coal and rock identification. This method includes optimizing and modifying the tunneling and anchoring machine equipment structure, real-time acquisition of multi-source operating condition data, coal and rock feature extraction based on DenseNet, coal and rock hardness quantification based on Gaussian process regression, coal and rock zoned identification based on an expert rule base, upper-computer waveform visualization and historical data storage, adaptive cutting control for coal and rock zones, and adaptive anchoring control for coal and rock zones. The specific technical solution is as follows:

[0126] (1) Structural optimization and transformation of the roadheader (core of speed improvement, efficiency enhancement, and stable operation):

[0127] To address the issues of insufficient power, severe wear of components such as cutting teeth, and machine vibration and deformation during heavy-load tunneling in conventional roadheader and anchor machine (BOM) for hard rock tunneling, this invention optimizes the overall power and load-bearing structure of the BOM, improving the overall operational stability from the hardware perspective and reducing mechanical wear. The main optimizations include:

[0128] Dual-motor coordinated power system: The system retains the original main cutting motor and symmetrically adds auxiliary cutting motors on the other side of the cutting drum. In actual operation, the main motor will always participate in the cutting work, but the auxiliary motor will only participate in rock breaking work to varying degrees in the hard rock area and the coal-rock mixed area. In order to adjust the working power of the two motors in real time to adapt to the cutting requirements of different zones of the cutting face, each motor is equipped with an independent frequency converter control module. In addition, the drum speed and the overall machine feed parameters will also be adjusted accordingly according to the different zones of the cutting face. Field tests show that the rock breaking ability of the roadheader of this invention is significantly improved in hard rock, and mechanical wear is reduced.

[0129] Reinforcement of key load-bearing structures of the whole machine: In order to solve the problem that the main beam of the machine body, the base of the cutting arm, the anchoring frame and the traveling load-bearing frame are prone to deformation when cutting high-hardness rocks;

[0130] The invention incorporates the following structural optimizations: high-strength, wear-resistant alloy steel plates are used to thicken and weld key stress-bearing locations; triangular reinforcing ribs are added to stress concentration areas; the core connection joint is replaced with an integrally cast structure, eliminating assembly gaps and potential stress concentration issues; damping and shock-absorbing components are added to the motor connection points and the cutting arm hinge points to reduce the resonance effects caused by the synchronous operation of dual motors. After the modifications, the overall structural rigidity is significantly improved, and vibration and deformation during rock cutting are noticeably reduced.

[0131] (2) Real-time acquisition and synchronization of multi-source operating condition data:

[0132] This invention requires the following three sensors to collect data simultaneously:

[0133] 1) Cutting surface image: A real-time visible light image of the cutting surface is captured by an explosion-proof industrial camera mounted on the cutting arm, denoted as:

[0134] ;

[0135] In the formula, This is a single-frame cut surface image data; The real number field indicates that each element in the matrix is ​​a real number; This represents the height of the image, i.e., the number of pixel rows in the vertical direction. 3 represents the width of the image, i.e., the number of pixel columns in the horizontal direction; 3 represents the number of channels of the image, corresponding to the three color channels of a color image: red (R), green (G), and blue (B).

[0136] 2) Cutting motor vibration data: Vibration acceleration of the cutting motor casing is collected using a triaxial accelerometer, with a sampling frequency of [missing information]. This constitutes a three-channel timing signal. :

[0137] ;

[0138] ;

[0139] In the formula, For time The vibration signal matrix is ​​the independent variable. In practical processing, this signal is a segment of a continuous time-domain signal. 3 represents the real number field; 3 represents the number of channels, corresponding to the vibration components in the three mutually perpendicular directions of X, Y, and Z collected by the triaxial accelerometer (installed on the flange edge of the front cover of the cutting motor); This represents the number of sampling points contained in the vibration signal segment. The sampling frequency of the vibration signal; The duration of the segment is in seconds.

[0140] 3) Cutting motor current data: The effective values ​​of the three-phase current of the cutting motor are obtained through Hall current sensors to form a single-channel timing signal. :

[0141] ;

[0142] ;

[0143] In the formula, For time The current signal vector is the independent variable, and a time segment synchronized with the vibration signal is extracted. For the real number field; This represents the number of sampling points contained in the current signal segment. The sampling frequency of the current signal; The duration of the segment is in seconds.

[0144] All three signals were synchronized in time according to the rotation cycle of the cutting drum.

[0145] (3) Multimodal feature extraction:

[0146] 1) Image Branch: Deep Feature Reuse Network (Improved DenseNet Structure):

[0147] Construct a densely connected convolutional network, whose basic units are composed of stacked feature reuse modules. The output of the feature multiplexing module Defined as:

[0148] ;

[0149] In the formula, This is a hierarchical index for the feature reuse module. =1,2,…, , This represents the total number of layers in the module. The initial input feature map for the network is the cut surface image. The feature tensor obtained after initial convolution (7×7 convolution, batch normalization, activation, and pooling); For the first The feature map tensor output by the layer feature reuse module is the feature map tensor of that layer after passing through a composite function. The new feature representation generated after the transformation; For the first The feature map output from the -1 layer feature reuse module is concatenated with the outputs of earlier layers and fed into the ... layer.

[0150] This network outputs the feature vector of the image. :

[0151] ;

[0152] In the formula, For the process The feature map output by the last layer of the network after the feature reuse module; is the dimension of the image feature vector.

[0153] 2) Temporal signal branch: Multi-scale one-dimensional residual network:

[0154] Vibration signals and current signal Robust features are extracted using multi-scale one-dimensional convolution modules.

[0155] The vibration signal is first processed by short-time energy normalization and low-pass filtering, then input into three layers of residual convolutional blocks, each containing parallel 1×3, 1×5, and 1×7 convolutional kernels. Multi-scale information is then fused through gated linear units to output a vibration feature vector. ,in The dimension of the vibration feature vector is determined by the number of output channels of the last convolutional layer.

[0156] The current signal undergoes similar processing to obtain the current feature vector. ,in This represents the dimension of the current feature vector. To preserve the statistical properties of the amplitude domain, manual features are also calculated. :

[0157] ;

[0158] In the formula, The kurtosis of the vibration signal; The root mean square of the vibration signal; The fluctuation rate of the current signal; The number of handcrafted features; This indicates transpose.

[0159] 3) Multimodal fusion features:

[0160] The above features are concatenated to form a comprehensive feature vector. :

[0161] ;

[0162] In the formula, The total dimension of the fused features numerically satisfies = .

[0163] (4) Numerical regression of hardness based on Gaussian process:

[0164] To obtain a numerical prediction of actual hardness that can express uncertainty, this invention uses a Gaussian process regression (GPR) model, whose input is the fused features. The output is the coal and rock hardness value. The specific process is as follows:

[0165] 1) Gaussian process modeling:

[0166] The hardness value follows a Gaussian prior process:

[0167] ;

[0168] In the formula, For different fusion characteristics; For fusion features The corresponding hardness value.

[0169] Zero-mean function The kernel function is a composite kernel capable of capturing the nonlinear effects of different modes. The composite nucleus includes the radial basis nucleus. polynomial kernel and periodic nuclei .

[0170] ;

[0171] ;

[0172] ;

[0173] ;

[0174] In the formula, For different fusion characteristics; for Transpose of; This serves as the variance scale for the radial basis kernel, controlling the amplitude of this kernel component; is the variance scale of the polynomial kernel; The variance scale for the periodic kernel; For periodic parameters, it represents the characteristic period of the load fluctuation of the cutting drum (in units of characteristic spatial distance). The length scale parameter of the radial basis kernel determines the smoothness of the function and the rate at which the correlation in the input space decays; The length scale parameter of the periodic kernel affects the smoothness of the periodic components; For the constant bias term of the polynomial kernel, This controls the weights of lower-order interactions; The order of the polynomial kernel is a positive integer that determines the complexity of the nonlinear interaction.

[0175] The set of hyperparameters in this composite kernel and observation noise variance All through the maximum

[0176] 2) Training and Prediction:

[0177] The collected calibration dataset Input, where For the first The fusion features of each training sample This represents the actual hardness measured in the laboratory using samples taken on-site. The logarithmic marginal likelihood is:

[0178] ;

[0179] ;

[0180] ;

[0181] ;

[0182] Iterative adjustment via gradient optimizer and Until convergence.

[0183] For real-time new samples The predicted average hardness value and variance for:

[0184] ;

[0185] ;

[0186] This yields the predicted hardness value. and forecast uncertainty .

[0187] In the formula, This is a column vector composed of training hardness values; The total number of training samples; This is a set of hyperparameters (described above); The input matrix; The matrix of the composite kernel for training input, i.e. ,for Symmetric positive definite matrix; Let V be the variance of the Gaussian observation noise.

[0188] (5) The expert rule base's division of coal and rock zones:

[0189] The input variables for the expert rule base include:

[0190] Hardness prediction mean ;

[0191] Hardness prediction standard deviation , indicating uncertainty;

[0192] Visual proportion of coal and rock The category activation map (CAM) output by the dense network calculates the proportion of coal-bearing regions in the image. ;

[0193] Vibration Amplitude Index , indicating the cutting impact strength;

[0194] Current fluctuation index .

[0195] The output of the expert rule base is the partitioning results and confidence scores. The zoning results include: pure coal zone, soft rock zone, hard rock zone, and mixed coal-rock zone. The specific zoning process is as follows:

[0196] 1) Input fuzzification:

[0197] Define linguistic values ​​and membership functions for each variable: The language values ​​are low, low-medium, medium, high-medium, and high. The language value is small, medium, or large; The language values ​​are pure coal (>0.9), coal with little rock, rock with little coal, and hard rock (<0.1); The language value is weak, medium, or strong; The language value is stable, moderately fluctuating, or drastically fluctuating.

[0198] 2) Expert Rule Base:

[0199] There are a total of 35 rules, but only 6 of them are shown here, as follows:

[0200] Rule 1 (R1): If For low, For pure coal, For the weak, If the condition is stable, it is considered a pure coal zone (confidence level 0.95).

[0201] Rule 2 (R2): If For medium and low It is characterized by abundant coal and scarce rock. For China If the fluctuation is moderate, it indicates a coal-rock mixed zone (confidence level 0.85).

[0202] Rule 3 (R3): If For the middle and If it is large, then further observation is needed. ,like If there is more rock than coal, it is a coal-rock mixed zone. If the trend is towards hard rock, then it leans towards soft rock (considering that moderate hardness may be a mixture of soft rock and coal).

[0203] Rule 4 (R4): If For middle and high school It is characterized by abundant rocks and scarce coal. For strength, If the fluctuations are violent, it indicates a soft rock zone (the soft rock is broken, the vibrations are large but the hardness is not high).

[0204] Rule 5 (R5): If For high, For hard rock, For strength, If the fluctuation is drastic, it indicates a hard rock zone (confidence level 0.95).

[0205] Rule 6 (R6): When When uncertainty is high, the overall confidence level of partitioning decisions decreases, and the decision relies more heavily on... and The strong evidence rule shall be used to make the decision.

[0206] 3) Fuzzy reasoning and defuzzification:

[0207] The Mamdani minimum inference method and the centroid method are used to defuzzify the data, outputting continuous partition membership scores. Finally, based on the maximum membership principle, the partition categories are determined: pure coal zone, soft rock zone, mixed coal-rock zone, and hard rock zone. The confidence score for each category is also output. .

[0208] (6) Real-time visualization of waveforms and storage of historical data on the host computer:

[0209] The host computer can display real-time data such as vibration waveform, current waveform, feed pressure, current zone results, and hardness for each motor, which on-site maintenance personnel can directly view. The system also saves this data to a local database and the cloud, facilitating staff to investigate the cause of faults and review the excavation and anchoring conditions.

[0210] (7) Adaptive cutting control for coal and rock zones:

[0211] Based on the cutting surface partitioning results output by the DenseNet network and the real-time coal and rock hardness values ​​obtained from Gaussian process regression, combined with the modified dual-motor cooperative power structure, different tunneling power and operating parameters are used. The specific partitioning and cutting control methods are as follows:

[0212] 1) Pure coal zone: Low hardness and low tunneling resistance, the system adopts a high-speed, low-load cutting mode. Only the main cutting motor is turned on to operate independently, while the auxiliary cutting motor enters a sleep standby state to avoid unnecessary energy waste caused by the simultaneous operation of two motors; the drum rotation speed and the overall machine feed speed are appropriately increased to reduce the single cutting depth.

[0213] 2) Soft rock area: The rock mass is loose and easily broken and collapses. The system controls two cutting motors to operate in coordination at low power, while slightly reducing the feed speed of the machine to prevent the surrounding rock from collapsing due to excessive feed. At the same time, it avoids the problem of over-excavation of the cross section and ensures the flatness of the roadway.

[0214] 3) Hard rock areas: High tunneling resistance and a tendency to jam are common. The system controls dual motors to operate at full power synchronously, while simultaneously reducing the overall machine feed speed and cutting depth per pass. Combined with the previously reinforced machine body, this fully leverages the high-power rock-breaking advantage of the dual motors, effectively solving on-site problems such as insufficient power from a single motor, jamming during hard rock tunneling, and excessive wear of the cutting teeth.

[0215] 4) Coal-rock mixed zone: The hardness of different areas may be different, and different hardness corresponds to different cutting parameters. However, the switching of cutting parameters should be gradual to avoid violent vibration of the machine body causing rock wall fracture.

[0216] (8) Real-time adaptive anchoring control based on coal and rock hardness:

[0217] The cutting surface is divided into high-hardness rock zone, low-hardness coal zone, low-hardness soft rock zone, and medium-hardness coal-rock mixed zone. Different anchoring operation parameters are used for different zones, including:

[0218] 1) High-hardness hard rock areas: Appropriately increase the anchoring drilling speed and increase the drilling feed pressure, while simultaneously increasing the anchoring depth and anchor bolt locking torque. This area has good overall integrity, but the drilling engagement is difficult. By adjusting the anchoring parameters, it is possible to ensure that the anchor bolt and the rock strata are fully engaged, avoiding safety problems such as anchor bolt loosening and support structure detachment in the later stages.

[0219] 2) Low-hardness coal and soft rock areas: Reduce drilling speed and feed pressure, decrease anchor bolt locking torque, and shorten anchoring depth. In these areas, the rock strength is low, and excessive anchoring force can easily cause borehole wall collapse, cracking of the surrounding rock, or excessive anchor bolt stress tearing the coal and rock mass.

[0220] 3) Coal-rock mixed zone: Instead of using uniform anchoring parameters, corresponding anchoring parameters are allocated according to the local real-time hardness of the cut surface to achieve precise anchoring in different zones.

[0221] Furthermore, in order to prevent coal mine disasters caused by excessively rapid tunneling, this system will dynamically limit the maximum feed rate based on geological conditions and safety thresholds to maintain safe tunneling.

[0222] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for zoned cutting control of roadheader based on coal and rock identification, characterized in that: The control steps include the following: Step S1: On the basis of the original main cutting motor of the roadheader, an auxiliary cutting motor is set up, and the dual-motor frequency conversion control system is installed and debugged; the main beam of the roadheader body, the cutting arm base, and the anchoring frame are reinforced by welding, strengthening with reinforcing ribs, and damping shock absorption components are installed. Step S2: During underground tunneling and anchoring operations, images of the coal and rock cutting face in the roadway are collected by explosion-proof industrial cameras deployed underground. Vibration signals, working current signals, and anchoring borehole pressure signals of the dual motors are collected synchronously by vibration sensors, current transformers, and pressure sensors deployed underground. After filtering and noise reduction preprocessing by the lower computer, the signals are uploaded to the upper computer. Step S3: Multimodal feature extraction is performed on the cut surface image through a deep feature reuse network. The vibration signal and operating current signal of the motor are obtained using a multi-scale one-dimensional residual network. The data are then fused by the host computer to generate fused features. Step S4: Input the fused features generated in step S3 into the Gaussian process regression model, and combine them with the calibration dataset to output the hardness value and confidence level of each coal and rock partition in real time. Step S5: Establish the judgment rules for pure coal area, soft rock area, hard rock area and coal-rock mixed area as an expert rule base, and then divide the cutting surface into pure coal area, soft rock area, hard rock area and coal-rock mixed area according to the output of step S3 and step S4. Step S6: The host computer displays the working current waveform of the dual motors, vibration signal, anchoring drilling pressure, current zone and hardness, and current excavation and anchoring parameters in real time. Each data is bound with a timestamp and stored in the local database and the cloud. Step S7: Based on the coal and rock zoning results obtained in step S5, adaptively adjust the output power, rotation speed, feed speed, and single cutting depth of the dual motors to achieve differentiated cutting control for different geological zones; Step S8: Based on the real-time detection results of coal and rock zoning, adaptively adjust the drilling speed, drilling pressure, anchoring depth, and anchor bolt torque of the anchoring operation to achieve differentiated anchoring control for different geological zones.

2. The method for zoned cutting control of roadheader based on coal and rock identification according to claim 1, characterized in that: The specific method for step S1 is as follows: The auxiliary cutting motor and the main cutting motor are symmetrically arranged on both sides of the cutting drum. The main cutting motor is used for the main cutting operation of coal and rock, and the auxiliary cutting motor is used for auxiliary cutting in the rocky area. The two motors are equipped with independent frequency conversion control modules, and the output parameters are adjusted independently according to the coal and rock zones.

3. The method for zoned cutting control of roadheader based on coal and rock identification according to claim 2, characterized in that: The specific method for step S2 is as follows: The cut surface image is specifically captured by an explosion-proof industrial camera mounted on the cutting arm, acquiring real-time visible light image data of the cut surface, expressed as: ; In the formula, This is a single-frame cut surface image data; The real number field indicates that each element in the matrix is ​​a real number; The height of the image. 3 represents the width of the image, and 3 represents the number of channels in the image, corresponding to the red, green, and blue color channels of a color image. The vibration signal is specifically acquired using a triaxial accelerometer to collect the vibration acceleration during the cutting of the motor housing, with a sampling frequency of... This constitutes a three-channel timing signal. The expression is: ; ; In the formula, For time The vibration signal matrix is ​​the independent variable; 3 represents the real number field; 3 represents the number of channels, corresponding to the vibration components in the three mutually perpendicular directions of X, Y, and Z acquired by the triaxial accelerometer. This represents the number of sampling points contained in the vibration signal segment. The sampling frequency of the vibration signal; This refers to the duration of the segment; The operating current signal is specifically acquired through a Hall current sensor. By collecting the effective values ​​of the three-phase current of the cutting motor, a single-channel timing signal is formed. The expression is: ; ; In the formula, For time The current signal vector is the independent variable, and a time segment synchronized with the vibration signal is extracted. For the real number field; This represents the number of sampling points contained in the current signal segment. The sampling frequency of the current signal; This refers to the duration of the segment.

4. The method for zoned cutting control of roadheader based on coal and rock identification according to claim 3, characterized in that: The specific method for step S3 is as follows: Multimodal feature extraction of the cut surface image is performed using the DenseNet network, including: Construct a densely connected convolutional network. The basic unit of the convolutional network is composed of several feature reuse modules stacked together. The output of the feature multiplexing module Defined as: ; In the formula, This is a hierarchical index for the feature reuse module. =1,2,…, , This represents the total number of layers in the module. The initial input feature map for the network is the cut surface image. The feature tensor obtained after initial convolution; For the first The feature map tensor output by the layer feature reuse module is the feature map tensor of that layer after passing through a composite function. The new feature representation generated after the transformation; For the first The feature map output from the -1 layer feature reuse module is concatenated with the outputs of earlier layers and fed into the ... layer; The feature vector of the output image of the convolutional network The expression is: ; In the formula, For the process The feature map output by the last layer of the network after the feature reuse module; The dimension of the image feature vector; The vibration signal and operating current signal of the motor are obtained using a multi-scale one-dimensional residual network, including: Vibration signals and current signal Robust features are extracted using multi-scale one-dimensional convolutional modules, where: Vibration signal First, the data undergoes short-time energy normalization and low-pass filtering. Then, it is input into three layers of residual convolutional blocks, each containing parallel 1×3, 1×5, and 1×7 convolutional kernels. Finally, multi-scale information is fused through gated linear units to output a vibration feature vector. ,in The dimension of the vibration feature vector is determined by the number of output channels of the last convolutional layer; Current signal After processing, the current characteristic vector is obtained. ,in The dimension of the current feature vector is used to preserve the statistical properties of the amplitude domain, while also calculating handcrafted features. : ; In the formula, The kurtosis of the vibration signal; The root mean square of the vibration signal; The fluctuation rate of the current signal; The number of handcrafted features; Indicates transpose; The various data will be merged, including: The above features are concatenated to form a comprehensive feature vector. : ; In the formula, The total dimension of the fused features numerically satisfies = .

5. The method for zoned cutting control of roadheader based on coal and rock identification according to claim 4, characterized in that: The specific method for step S4 is as follows: A Gaussian process regression model was used, with the fused features as input. The output is the coal and rock hardness value. This yields a numerical prediction of actual hardness with the ability to express uncertainty, including: Modeling a Gaussian process: The hardness value follows a Gaussian prior process, expressed as: ; In the formula, For different fusion characteristics; For fusion features The corresponding hardness value; Zero-mean function The kernel function is a composite kernel capable of capturing the nonlinear effects of different modes. Including radial base core polynomial kernel Periodic nuclei The expression is: ; ; ; ; In the formula, For different fusion characteristics; for Transpose of; This serves as the variance scale for the radial basis kernel, controlling the amplitude of this kernel component; is the variance scale of the polynomial kernel; The variance scale for the periodic kernel; The periodic parameter represents the characteristic period of the load fluctuation of the cutting drum; The length scale parameter of the radial basis kernel determines the smoothness of the function and the rate at which the correlation in the input space decays; The length scale parameter of the periodic kernel affects the smoothness of the periodic components; For the constant bias term of the polynomial kernel, This controls the weights of lower-order interactions; The order of the polynomial kernel is a positive integer, which determines the complexity of the nonlinear interaction. Training and prediction include: The collected calibration dataset Input, where For the first The fusion features of each training sample The logarithmic marginal likelihood is the true hardness measured in the laboratory using on-site sampling: ; ; ; ; Iterative adjustment via gradient optimizer and Until convergence; For real-time new samples The predicted average hardness value and variance for: ; ; This yields the predicted hardness value. and forecast uncertainty ; In the formula, This is a column vector composed of training hardness values; The total number of training samples; For hyperparameter set; The input matrix; The matrix of the composite kernel for training input, i.e. ,for Symmetric positive definite matrix; Let V be the variance of the Gaussian observation noise.

6. The method for zoned cutting control of roadheader based on coal and rock identification according to claim 5, characterized in that: The specific method for step S5 is as follows: The input variables for defining an expert rule base include: Hardness prediction mean ; Hardness prediction standard deviation , indicating uncertainty; Visual proportion of coal and rock The category activation map output by the dense network calculates the proportion of coal-bearing regions in the image. ; Vibration Amplitude Index , indicating the cutting impact strength; Current fluctuation index ; The output of the expert rule base is the partitioning results and confidence scores. The zoning results include: pure coal zone, soft rock zone, hard rock zone, and mixed coal-rock zone. The zoning method is as follows: Define linguistic values ​​and membership functions for each variable: The language values ​​are low, low-medium, medium, high-medium, and high. The language value is small, medium, or large; The language values ​​are: pure coal, coal with little rock, rock with little coal, hard rock; The language value is weak, medium, or strong; The language value is stable, with moderate fluctuations, or with drastic fluctuations. Define the expert rule base rules as follows: Rule 1: If For low, For pure coal, For the weak, For stability, it is a pure coal zone; Rule 2: If For medium and low It is characterized by abundant coal and scarce rock. For China If the fluctuation is moderate, it indicates a coal-rock mixture zone; Rule 3: If For the middle and If it is large, then further observation is needed. ,like If there is more rock than coal, it is a coal-rock mixed zone. If the tendency is towards hard rock, then it leans towards soft rock areas; Rule 4: If For middle and high school It is characterized by abundant rocks and scarce coal. For strength, If the fluctuations are violent, it indicates a soft rock area; Rule 5: If For high, For hard rock, For strength, If the fluctuations are violent, it indicates a hard rock zone; Rule 6: When When the value is large, the overall confidence level of partition determination decreases, and it relies more on... and The strong evidence rule shall be used to make the decision; Performing fuzzy reasoning and defuzzification: The Mamdani minimum inference method and the centroid method are used to defuzzify the data, outputting continuous partition membership scores. Finally, based on the maximum membership principle, the partition categories are determined: pure coal zone, soft rock zone, coal-rock mixed zone, and hard rock zone, and the confidence score of each category is also output. .

7. The method for zoned cutting control of roadheader based on coal and rock identification according to claim 6, characterized in that: The specific method for step S7 is as follows: Based on the cutting surface partitioning results output by the DenseNet network and the real-time coal and rock hardness values ​​obtained from Gaussian process regression, combined with the modified dual-motor cooperative power structure, corresponding tunneling power and operating parameters are used for partitioned cutting control, including: For pure coal areas: the system only turns on the main cutting motor to operate independently, while the auxiliary cutting motor enters a sleep standby state, increasing the drum rotation speed and the overall machine feed speed, and reducing the single cutting depth; For soft rock areas: the system controls two cutting motors to operate in coordination at low power, while adjusting the machine feed speed to prevent excessive feed from causing local collapse of the surrounding rock, avoid over-excavation of the cross section, and ensure the flatness of the roadway formation; For hard rock areas: The system controls dual motors to output full power synchronously, reducing the overall machine feed speed and increasing the single cutting depth; For coal-rock mixed zones: different cutting parameters are used for different hardness levels.

8. The method for zoned cutting control of roadheader based on coal and rock identification according to claim 7, characterized in that: The specific method for step S8 is as follows: An adaptive anchoring control scheme is adopted, dividing the cutting surface into high-hardness rock zone, low-hardness coal zone, low-hardness soft rock zone, and medium-hardness coal-rock mixed zone. The following anchoring operation parameters are used for control of different zones: For high-hardness hard rock areas: increase the anchoring drilling speed, increase the drilling feed pressure, and simultaneously increase the anchoring depth and anchor bolt locking torque; For low-hardness coal areas and soft rock areas: reduce drilling speed and feed pressure, reduce anchor bolt locking torque, and shorten anchoring depth; For coal-rock mixed zones: instead of using uniform anchoring parameters, corresponding anchoring parameters are assigned according to the local real-time hardness of the cut surface.