An Automatic Correction Method for Roadheader and Anchor Machines Based on Multi-Source Fusion Positioning

Abstract

This invention relates to the field of control technology for intelligent underground tunneling equipment, and more specifically to an automatic deviation correction method for roadheader-anchor machine (BAM) based on multi-source fusion positioning. By constructing a digital twin incorporating the dynamic characteristics of the BAM and the interaction characteristics between the surrounding rock and the equipment, an absolute positioning map based on the inherent characteristics of the tunnel's surrounding rock is established. A federated filtering architecture is used to fuse multi-source heterogeneous data, achieving autonomous positioning without external references and fault-tolerant reconstruction for sensor failures. Combining tunneling trend prediction and multi-degree-of-freedom collaborative control, an active control system of "pre-correction + real-time correction + post-compensation" is constructed. This invention completely eliminates dependence on external references, maintains stable correction accuracy under various complex working conditions, enables ultra-long-distance tunnel excavation, and significantly improves tunnel formation quality and tunneling efficiency.

Description

Technical Field

[0001] This invention relates to the field of control technology for intelligent underground tunneling equipment, and more specifically to an automatic deviation correction method for tunneling and anchoring machines based on multi-source fusion positioning. Background Technology

[0002] Roadheaders are the core equipment for tunneling underground in coal mines. The accuracy of their tunneling direction directly affects the quality of tunnel formation, construction safety, and subsequent mining efficiency. Currently, the main bottlenecks in automatic deviation correction technology for roadheader and anchor machine (BOM) are as follows: Existing high-precision positioning technologies all require the pre-deployment of external equipment such as laser pointers and UWB base stations, which necessitate frequent station relocations as tunneling progresses. Furthermore, these systems are susceptible to damage from underground collapses and equipment collisions, leading to system interruptions and preventing continuous tunneling. Most existing technologies employ Kalman filtering and its variants, which have limited capabilities in handling non-Gaussian and non-stationary noise, and lack the ability to autonomously diagnose and reconstruct sensor faults; a single sensor failure can paralyze the system. Existing technologies operate on a passive "correction after deviation occurs" model, failing to predict tunneling trends and resulting in "snake-like" tunneling during the correction process, leading to poor tunnel formation quality and severe over- and under-excavation. Under complex geological conditions such as hard rock, fractured zones, and faults, the attitude disturbances caused by sudden changes in the stress on the BOM cannot be effectively suppressed, causing a sharp decline in correction accuracy and even preventing automatic tunneling. As the tunneling distance increases, the cumulative error effect of the inertial navigation system becomes significant, making it difficult for existing technologies to meet the accuracy requirements of long-distance tunneling.

[0003] Therefore, how to provide an automatic deviation correction method for tunneling and anchoring machines based on multi-source fusion positioning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, the present invention provides an automatic deviation correction method for tunneling and anchoring machines based on multi-source fusion positioning, which aims to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An automatic deviation correction method for roadheader / anchor machine based on multi-source fusion positioning includes the following steps: S1. Construct a digital twin that includes the dynamic characteristics of the tunneling and anchoring machine and the interaction characteristics between the surrounding rock and the equipment, complete the initial system calibration, and construct the initial surrounding rock feature map; S2. Synchronously collect the operating status data of the tunneling and anchoring machine and the surrounding rock perception data of the roadway, and extract the stability characteristics of the surrounding rock surface; S3. Employ a multi-source heterogeneous data fusion positioning algorithm to calculate the real-time six-degree-of-freedom pose information of the tunneling and anchoring machine, thereby enabling autonomous diagnosis of sensor faults and fault-tolerant system reconfiguration. S4. Based on the digital twin and the tunneling trend prediction model, predict the position change trend of the tunneling and anchoring machine within a preset number of tunneling cycles in the future. When the predicted deviation exceeds the preset threshold, generate a pre-correction control command and execute it before the deviation actually occurs. S5. A multi-degree-of-freedom collaborative control algorithm is adopted to simultaneously optimize the operating parameters of the walking mechanism, cutting mechanism and support mechanism, so as to realize the real-time correction control of the tunneling and anchoring machine, and introduce equipment state compensation to improve the correction accuracy. S6. During the tunneling process, new roadway surrounding rock perception data are collected periodically, the global surrounding rock feature map is updated, the surrounding rock feature matching results are used to make absolute correction of the cumulative positioning error, and the model parameters of the digital twin are updated at the same time. S7. Real-time monitoring of surrounding rock conditions and equipment operating status; automatic adjustment of system control strategies according to different working conditions; and graded handling of abnormal situations.

[0006] Furthermore, the digital twin in S1 includes at least one or more of the following: fuselage multibody dynamics model, cutting mechanism load model, walking mechanism motion model, and surrounding rock-equipment interaction model; the initial surrounding rock feature map is constructed by using a three-dimensional sensing device to scan the surface of the surrounding rock of the tunnel within a preset range of the tunneling starting section, extracting multiple stable geometric feature points and adding corresponding semantic labels.

[0007] Furthermore, the operating status data of the tunneling and anchoring machine in S2 includes one or more of the following: inertial navigation data, odometer data, hydraulic system parameters, cutting mechanism load data, and machine body deformation data; the roadway surrounding rock perception data includes one or more of the following: three-dimensional point cloud data, surrounding rock hardness data, and surrounding rock structure data.

[0008] Furthermore, the multi-source heterogeneous data fusion positioning algorithm in S3 adopts a federated filtering architecture, which includes multiple independent local filters and a master filter; the multiple local filters include at least a high-frequency pose prediction filter that fuses inertial navigation and odometry data, an absolute pose correction filter that fuses three-dimensional perception data and surrounding rock feature matching data, and a pose compensation filter based on fuselage deformation data.

[0009] Furthermore, the sensor fault autonomous diagnosis in S3 adopts the residual test method. When the residual of the local filter corresponding to a certain sensor exceeds the preset threshold, the sensor is determined to be faulty. The main filter automatically adjusts the fusion weight of each local filter to realize the fault-tolerant reconstruction of the system.

[0010] Furthermore, the tunneling trend prediction model in S4 adopts a deep learning model, with real-time pose data, equipment operation data and surrounding rock characteristic parameters as inputs, and the output being the predicted pose values ​​of the tunneling machine within a preset number of tunneling cycles in the future; the pre-correction control command is generated based on fuzzy reasoning or rule reasoning, and the pre-correction action is executed after the current tunneling cycle ends and before the next cycle begins.

[0011] Furthermore, the multi-degree-of-freedom collaborative control algorithm in S5 adopts a model predictive control algorithm, with the objective function being to minimize the deviation and control quantity change within the future preset control period, while simultaneously optimizing one or more of the left and right track speeds, cutting arm posture parameters, and multiple support cylinder pressures; the equipment state compensation includes at least one or more of the following: body deformation compensation and track slippage compensation.

[0012] Furthermore, the global surrounding rock feature map update triggering conditions in S6 include the tunneling distance reaching a preset value or the positioning accuracy being lower than a preset threshold; the map update adopts an incremental SLAM algorithm, scanning the roadway surrounding rock surface within a preset range before and after the current position, extracting new surrounding rock features and matching and fusing them with the existing map.

[0013] Furthermore, the working condition identification in S7 uses a machine learning algorithm to classify the working conditions into one or more of the following: normal geology, hard rock, fractured zone, fault, and equipment abnormality. For different working conditions, one or more of the following are adjusted: tunneling speed, map update interval, control frequency, and equipment operating parameters. The abnormal situation classification and handling includes automatic switching of fusion mode and alarm for sensor failure, automatic speed reduction or shutdown and alarm for equipment abnormality, and immediate cessation of all actions and disconnection of power supply in emergency situations.

[0014] Compared to existing technologies, this invention proposes an absolute positioning method based on the inherent characteristics of the surrounding rock of the tunnel. It eliminates the need for any pre-installed external equipment, achieving truly autonomous tunneling and solving the problems of frequent relocation and easy damage of external reference points. It transforms the traditional "passive correction" into an active control mode of "pre-correction + real-time correction + post-compensation," effectively avoiding "snake-like" tunneling and significantly improving tunnel formation quality. Employing a federated filtering architecture, it possesses autonomous fault diagnosis and reconstruction capabilities for sensors; single sensor failure does not affect normal system operation, significantly extending the system's mean time between failures (MTBF). By simulating the response of the roadheader under different geological conditions using a digital twin, combined with an adaptive control strategy, it maintains stable correction accuracy under various complex working conditions. Periodic absolute correction using surrounding rock characteristics enables ultra-long-distance tunneling with accumulated errors controlled within acceptable limits. Without manual intervention or external reference point relocation, it achieves continuous automatic tunneling, significantly improving tunneling efficiency and reducing worker labor intensity and safety risks. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating an automatic deviation correction method for a tunneling and anchoring machine based on multi-source fusion positioning according to the present invention. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See appendix Figure 1 This invention discloses an automatic deviation correction method for tunneling and anchoring machines based on multi-source fusion positioning, comprising: S1. Construct a digital twin that includes the dynamic characteristics of the tunneling and anchoring machine and the interaction characteristics of the surrounding rock and equipment, complete the initial system calibration and construct an initial surrounding rock feature map; the digital twin includes at least one or more of the following: a multi-body dynamic model of the machine body, a load model of the cutting mechanism, a motion model of the traveling mechanism and a surrounding rock and equipment interaction model; the initial surrounding rock feature map is constructed by using a three-dimensional sensing device to scan the surface of the surrounding rock of the tunnel within a preset range at the starting section of the tunneling, extracting multiple stable geometric feature points and adding corresponding semantic labels.

[0019] S2. Synchronously collect the operating status data of the roadheader and the surrounding rock perception data, and extract the stability features of the surrounding rock surface; the operating status data of the roadheader includes one or more of the following: inertial navigation data, odometer data, hydraulic system parameters, cutting mechanism load data, and machine body deformation data; the surrounding rock perception data includes one or more of the following: three-dimensional point cloud data, surrounding rock hardness data, and surrounding rock structure data.

[0020] S3. A multi-source heterogeneous data fusion positioning algorithm is adopted to calculate the real-time six-degree-of-freedom pose information of the tunneling and anchoring machine, realizing autonomous sensor fault diagnosis and system fault-tolerant reconstruction. The multi-source heterogeneous data fusion positioning algorithm adopts a federated filtering architecture, which includes multiple independent local filters and a main filter. The multiple local filters include at least a high-frequency pose prediction filter that fuses inertial navigation and odometer data, an absolute pose correction filter that fuses three-dimensional perception data and surrounding rock feature matching data, and a pose compensation filter based on fuselage deformation data. The autonomous sensor fault diagnosis adopts the residual test method. When the residual of the local filter corresponding to a certain sensor exceeds a preset threshold, the sensor is determined to be faulty. The main filter automatically adjusts the fusion weights of each local filter to realize system fault-tolerant reconstruction.

[0021] S4. Based on the digital twin and the tunneling trend prediction model, predict the position change trend of the roadheader within a preset number of tunneling cycles. When the predicted deviation exceeds a preset threshold, generate a pre-correction control command and execute it before the deviation actually occurs. The tunneling trend prediction model adopts a deep learning model, inputting real-time position data, equipment operation data, and surrounding rock characteristic parameters, and outputting the predicted position value of the roadheader within a preset number of tunneling cycles. The pre-correction control command is generated based on fuzzy reasoning or rule reasoning, and the pre-correction action is executed after the current tunneling cycle ends and before the next cycle begins.

[0022] S5. A multi-degree-of-freedom collaborative control algorithm is adopted to simultaneously optimize the operating parameters of the walking mechanism, cutting mechanism, and support mechanism, thereby achieving real-time correction control of the tunneling and anchoring machine. Equipment state compensation is introduced to improve the correction accuracy. The multi-degree-of-freedom collaborative control algorithm adopts a model predictive control algorithm with the objective function of minimizing the deviation and control quantity change within the future preset control cycle. At the same time, it optimizes one or more of the following: left and right track speeds, cutting arm posture parameters, and pressure of multiple support cylinders. The equipment state compensation includes at least one or more of the following: body deformation compensation and track slippage compensation.

[0023] S6. During the tunneling process, new roadway surrounding rock perception data are collected periodically to update the global surrounding rock feature map. The accumulated positioning error is absolutely corrected using the surrounding rock feature matching results, and the model parameters of the digital twin are updated simultaneously. The global surrounding rock feature map update is triggered when the tunneling distance reaches a preset value or the positioning accuracy is lower than a preset threshold. The map update adopts an incremental SLAM algorithm to scan the roadway surrounding rock surface within a preset range before and after the current position, extract new surrounding rock features, and match and fuse them with the existing map.

[0024] S7. Real-time monitoring of surrounding rock conditions and equipment operating status; automatic adjustment of system control strategies based on different working conditions; graded handling of abnormal situations; working condition identification uses machine learning algorithms to classify working conditions into one or more of the following: normal geology, hard rock, fractured zone, fault, and equipment abnormality; and adjusts one or more of the following for different working conditions: tunneling speed, map update interval, control frequency, and equipment operating parameters; graded handling of abnormal situations includes automatic switching to fusion mode and alarm for sensor failure, automatic speed reduction or shutdown and alarm for equipment abnormality, and immediate cessation of all actions and disconnection of power supply in emergency situations.

[0025] This embodiment provides a preferred implementation of the present invention based on lidar and federated unscented Kalman filtering, and the specific steps are as follows: A digital twin of a roadheader is constructed in an edge computing unit, including a multibody dynamics model of the fuselage (24 degrees of freedom), a load model of the cutting mechanism (prediction error ≤5%), a slippage model of the traveling mechanism (prediction error ≤3%), and a rock-equipment interaction model (discrete element method). Establish a global coordinate system O-XYZ for the tunnel (origin at the starting point of the tunneling face, X-axis forward along the design centerline, Y-axis pointing to the right, and Z-axis vertically upward) and a coordinate system O'-X'Y'Z' for the roadheader body (origin at the center of rotation of the cutting arm, X' axis along the forward direction). The transformation matrix calibration between the sensor and the fuselage coordinate system was completed using the hand-eye calibration method, with a calibration accuracy better than 0.5 mm; the sensor was synchronized at the microsecond level using the IEEE 1588 PTP protocol, with a synchronization error ≤1 μs. At the starting position of the tunneling, a 3D lidar was used to scan the surrounding rock surface within a 20m range ahead. After statistical filtering (threshold is 3 times the standard deviation of the average distance), voxel downsampling (5cm×5cm×5cm), and ground segmentation, an improved SIFT algorithm was used to extract no less than 500 stable geometric feature points (joints, fissures, concave and convex surfaces, rock block boundaries) and add semantic labels to construct an initial surrounding rock feature map. The feature point repetition recognition rate was greater than 95%. Import the centerline and cross-sectional parameters of the tunnel design to generate a three-dimensional preset tunneling trajectory with a point spacing of 0.1m, which includes the target coordinates, attitude angle and cross-sectional profile of each location.

[0026] A high-precision strapdown inertial navigation system (gyroscope zero bias ≤0.01° / h, accelerometer zero bias ≤50μg, sampling frequency 100Hz) is installed at the center of the top of the fuselage; an explosion-proof three-dimensional lidar (360° horizontal field of view, -45°~+45° vertical field of view, ranging range 0.5~100m, sampling frequency 10Hz, ranging accuracy ±2cm) is installed on the upper front; distributed fiber optic strain sensors (total length 50m, spatial resolution 1m, sampling frequency 50Hz, strain accuracy ±1με) are arranged on the main beam, cutting arm, and track frame; each of the left and right track hydraulic motors is equipped with a pressure sensor (0~30MPa, 0.5 grade), a flow sensor (0~200L / min, 1 grade), and a speed sensor (0~1000rpm, 0.1 grade); the cutting motor is equipped with a current sensor (0~500A, 0.5 grade) and a torque sensor (…). (0.5 grade); each of the four support cylinders is equipped with a pressure sensor (0~30MPa, 0.5 grade); Inertial navigation data undergoes zero-bias correction, gravity compensation, and temperature compensation; lidar point cloud data undergoes statistical filtering for noise reduction, voxel downsampling, and ground segmentation; distributed fiber optic data undergoes temperature compensation and strain calculation; hydraulic and cutting data are filtered using a moving average (window size 5) to remove high-frequency noise. An improved Harris corner detection and region growing algorithm was used to extract corner, planar and curved surface features of the surrounding rock. A 128-dimensional feature descriptor was calculated, and erroneous matching points were removed by the Random Sampling Consensus (RANSAC) algorithm, with a matching accuracy of more than 90%.

[0027] Construct a three-layer federated unscented Kalman filter (FUKF) architecture, where the system state vector X=[x,y,z, , ,v_z,φ,θ,ψ, , [b_z]^T (position, velocity, attitude angle, gyroscope zero bias); The first local filter (100Hz) is used to fuse inertial navigation and odometry data, and employs the unscented Kalman filter (UKF) algorithm to provide high-frequency pose prediction; The second local filter (10Hz) is used to fuse lidar SLAM and surrounding rock feature matching data, and employs the Iterative Closest Point (ICP) algorithm to provide absolute pose correction. The third local filter (50Hz) is used to calculate the elastic deformation of the fuselage based on distributed fiber optic strain data, providing pose compensation; The main filter is used to dynamically adjust the weights based on the residuals of each local filter for global fusion; the smaller the residual, the greater the weight. The residual χ² test method (95% confidence level) is used to monitor the sensor status in real time. When the residual exceeds the threshold, the weight of the faulty sensor is automatically reset to zero and the system switches to the remaining sensor fusion mode. Under normal operating conditions, the positioning accuracy is ±20mm for planar position, ±30mm for elevation, and ±0.1° for attitude angle.

[0028] Real-time pose (6D), walking and cutting mechanism load data (4D), and surrounding rock characteristic parameters (2D) are input into the digital twin. An LSTM network is used to predict the pose change trend within the next 3 tunneling cycles (approximately 1.5m). The network structure is: 12 nodes in the input layer, 64 nodes in the hidden layer, and 6 nodes in the output layer (3D position + 3D attitude). The prediction time step is 0.1s, and the prediction accuracy is: planar position ±30mm and attitude angle ±0.2°. The deviation between the predicted pose and the preset trajectory is calculated. When the predicted lateral deviation is >30mm or the predicted yaw angle deviation is >0.3°, pre-correction control commands (left and right track speed difference, initial swing angle of the cutting arm, and adjustment amount of support cylinder preload) are generated based on fuzzy inference. The pre-correction action is executed after the end of the current tunneling cycle and before the start of the next cycle, with an execution time ≤2s. After the pre-correction is executed, the digital twin simulates the pre-correction effect. If it is not ideal, the pre-correction amount is readjusted, which can eliminate more than 70% of the future deviation.

[0029] Every 0.1 seconds, the lateral deviation Δy, yaw angle deviation Δψ, pitch angle deviation Δθ, and roll angle deviation Δφ of the actual position and orientation of the roadheader are calculated from the preset trajectory. A constrained model predictive control (MPC) algorithm is used, with 10 control cycles (1s) in the prediction time domain and 3 control cycles (0.3s) in the control time domain. The objective function is: (e is the deviation, Δu is the rate of change of the control quantity, Q and R are the weight matrices); left track speed v_l, right track speed v_r, left and right swing angle of the cutting arm α, lifting angle β, extension and retraction l, pressure of the four support cylinders p1-p4; The constraints are 0 ≤ v_l, v_r ≤ 1.5 m / s, |v_l - v_r| ≤ 0.5 m / s, -45° ≤ α ≤ +45°, -30° ≤ β ≤ +30°, and 5 MPa ≤ p1 - p4 ≤ 20 MPa. The fuselage deformation is compensated based on distributed optical fiber data (accuracy ±5mm), and track slippage is compensated based on hydraulic parameters and odometer data (accuracy ±2%). The optimization problem is solved using quadratic programming, and the control command for the first control cycle is executed. The real-time correction response time is ≤0.5s, and the correction accuracy is: lateral deviation ≤±20mm, yaw angle deviation ≤±0.1°.

[0030] Map updates are triggered automatically every 50m of tunneling, or immediately when the positioning accuracy is 50mm below the planar position. During updates, tunneling is stopped, and the surrounding rock surface within a 25m radius before and after the current position is scanned using a 3D LiDAR scanner. New surrounding rock features are extracted and matched with the existing map. An incremental SLAM algorithm is used to update the global feature map, with an update time of ≤30s. The absolute pose of the tunneling machine is calculated using the matching results, and the cumulative error of the inertial navigation system is corrected. After correction, the cumulative error of the inertial navigation system is ≤±30mm / 50m. The parameters of the digital twin, such as the hardness of the surrounding rock and the degree of joint development, are updated according to the new surrounding rock features. The dynamic model parameters are optimized once every 100m of tunneling using a reinforcement learning algorithm.

[0031] The system monitors surrounding rock conditions and equipment operating status in real time, employing a Support Vector Machine (SVM) algorithm for condition identification with an accuracy rate exceeding 95%. Adaptive control strategies include: normal geological mode (tunneling speed 1.0–1.5 m / min, map update interval 50 m); hard rock mode (surrounding rock hardness > 60 MPa, tunneling speed 0.5–0.8 m / min, map update interval 30 m); fractured zone mode (tunneling speed 0.3–0.5 m / min, map update interval 10 m, support cylinder preload 15 MPa); and fault mode (tunneling speed 0.2–0.3 m / min, map update interval 5 m, short advance, multi-cycle tunneling). In case of sensor failure, the system automatically switches to a fusion mode and issues an audible and visual alarm. When the equipment load exceeds 120% of its rated value, it automatically reduces speed; when it exceeds 150%, it automatically shuts down and issues an alarm. In emergency situations such as excessive gas levels or roof collapse, all operations are immediately stopped, power is cut off, and an emergency alarm is issued.

[0032] This embodiment provides another alternative implementation of the present invention based on binocular vision and extended Kalman filtering, the main difference from embodiment 1 being: An explosion-proof binocular vision camera (baseline distance 120mm, resolution 1280×720, sampling frequency 20Hz) was used to replace the 3D LiDAR. This camera acquired 3D point cloud data of the surrounding rock surface in the tunnel using stereo vision, with a ranging range of 0.5–50m and a ranging accuracy of ±3cm. A Federated Extended Kalman Filter (FEKF) was used instead of a Federated Unscented Kalman Filter, reducing computational complexity by approximately 40%. A GRU network was used instead of an LSTM network, reducing model parameters by approximately 30% and improving prediction speed by approximately 25%. An ORB algorithm was used instead of a SIFT algorithm, improving feature extraction speed by approximately 50%.

[0033] The other steps in this embodiment are basically the same as in Embodiment 1. While ensuring basic performance, the system cost and computational complexity are reduced, making it suitable for cost-sensitive small and medium-sized coal mine application scenarios.

[0034] The performance comparison data of this invention with existing mainstream roadheader automatic correction technology is shown in the table below:

[0035] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0036] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for automatic deviation correction of a roadheader / anchor machine based on multi-source fusion positioning, characterized in that, Includes the following steps: S1. Construct a digital twin that includes the dynamic characteristics of the tunneling and anchoring machine and the interaction characteristics between the surrounding rock and the equipment, complete the initial system calibration, and construct the initial surrounding rock feature map; S2. Synchronously collect the operating status data of the tunneling and anchoring machine and the surrounding rock perception data of the roadway, and extract the stability characteristics of the surrounding rock surface; S3. Employ a multi-source heterogeneous data fusion positioning algorithm to calculate the real-time six-degree-of-freedom pose information of the tunneling and anchoring machine, thereby enabling autonomous diagnosis of sensor faults and fault-tolerant system reconfiguration. S4. Based on the digital twin and the tunneling trend prediction model, predict the position change trend of the tunneling and anchoring machine within a preset number of tunneling cycles in the future. When the predicted deviation exceeds the preset threshold, generate a pre-correction control command and execute it before the deviation actually occurs. S5. A multi-degree-of-freedom collaborative control algorithm is adopted to simultaneously optimize the operating parameters of the walking mechanism, cutting mechanism and support mechanism, so as to realize the real-time correction control of the tunneling and anchoring machine, and introduce equipment state compensation to improve the correction accuracy. S6. During the tunneling process, new roadway surrounding rock perception data are collected periodically, the global surrounding rock feature map is updated, the surrounding rock feature matching results are used to make absolute correction of the cumulative positioning error, and the model parameters of the digital twin are updated at the same time. S7. Real-time monitoring of surrounding rock conditions and equipment operating status; automatic adjustment of system control strategies according to different working conditions; and graded handling of abnormal situations.

2. The automatic deviation correction method for a roadheader / anchor machine based on multi-source fusion positioning according to claim 1, characterized in that, The digital twin in S1 includes at least one or more of the following: fuselage multibody dynamics model, cutting mechanism load model, walking mechanism motion model, and surrounding rock-equipment interaction model; the initial surrounding rock feature map is constructed by using a three-dimensional sensing device to scan the surface of the surrounding rock of the tunnel within a preset range of the tunneling starting section, extracting multiple stable geometric feature points and adding corresponding semantic labels.

3. The automatic deviation correction method for a roadheader / anchor machine based on multi-source fusion positioning according to claim 1, characterized in that, The operating status data of the tunneling and anchoring machine in S2 includes one or more of the following: inertial navigation data, odometer data, hydraulic system parameters, cutting mechanism load data, and machine body deformation data; the roadway surrounding rock perception data includes one or more of the following: three-dimensional point cloud data, surrounding rock hardness data, and surrounding rock structure data.

4. The automatic deviation correction method for a roadheader / anchor machine based on multi-source fusion positioning according to claim 1, characterized in that, The multi-source heterogeneous data fusion positioning algorithm in S3 adopts a federated filtering architecture, which includes multiple independent local filters and a master filter. The multiple local filters include at least a high-frequency pose prediction filter that fuses inertial navigation and odometer data, an absolute pose correction filter that fuses three-dimensional perception data and surrounding rock feature matching data, and a pose compensation filter based on fuselage deformation data.

5. The automatic deviation correction method for a roadheader / anchor machine based on multi-source fusion positioning according to claim 1, characterized in that, The sensor fault autonomous diagnosis in S3 adopts the residual test method. When the residual of the local filter corresponding to a certain sensor exceeds the preset threshold, the sensor is determined to be faulty. The main filter automatically adjusts the fusion weight of each local filter to realize the fault-tolerant reconstruction of the system.

6. The automatic deviation correction method for a roadheader / anchor machine based on multi-source fusion positioning according to claim 1, characterized in that, The tunneling trend prediction model in S4 adopts a deep learning model. The inputs are real-time pose data, equipment operation data and surrounding rock characteristic parameters, and the output is the predicted pose value of the tunneling machine in the future preset number of tunneling cycles. The pre-correction control command is generated based on fuzzy reasoning or rule reasoning. The pre-correction action is executed after the current tunneling cycle ends and before the next cycle begins.

7. The automatic deviation correction method for a roadheader / anchor machine based on multi-source fusion positioning according to claim 1, characterized in that, The multi-degree-of-freedom collaborative control algorithm in S5 adopts a model predictive control algorithm, with the objective function being to minimize the deviation and control quantity change within the future preset control period. At the same time, it optimizes one or more of the left and right track speeds, cutting arm posture parameters, and multiple support cylinder pressures. The equipment state compensation includes at least one or more of the following: body deformation compensation and track slippage compensation.

8. The automatic deviation correction method for a roadheader / anchor machine based on multi-source fusion positioning according to claim 1, characterized in that, The global surrounding rock feature map update triggering conditions in S6 include the tunneling distance reaching a preset value or the positioning accuracy being lower than a preset threshold; the map update adopts an incremental SLAM algorithm, which scans the surrounding rock surface of the tunnel within a preset range before and after the current position, extracts new surrounding rock features, and matches and merges them with the existing map.

9. The automatic deviation correction method for a roadheader / anchor machine based on multi-source fusion positioning according to claim 1, characterized in that, The working condition identification described in S7 uses a machine learning algorithm to classify the working conditions into one or more of the following: normal geology, hard rock, fractured zone, fault, and equipment abnormality. For different working conditions, one or more of the following are adjusted: tunneling speed, map update interval, control frequency, and equipment operating parameters. The abnormal situation classification and handling includes automatic switching of fusion mode and alarm for sensor failure, automatic speed reduction or shutdown and alarm for equipment abnormality, and immediate cessation of all actions and disconnection of power supply in case of emergency.

Images