CN-122015866-A - Multi-sensor fusion self-adaptive cutting path planning method for coal mine tunneling robot

CN122015866ACN 122015866 ACN122015866 ACN 122015866ACN-122015866-A

Abstract

The invention provides a multi-sensor fusion self-adaptive cutting path planning method for a coal mine tunneling robot, and relates to the technical field of intelligent tunneling of coal mines. The coal mine tunneling robot multi-sensor fusion self-adaptive cutting path planning method comprises the steps of deploying a laser radar, a millimeter wave radar, an IMU, a coal and rock spectrum sensor and a vibration sensor five-source sensing system, precisely outputting coal and rock types, geological boundaries, machine body pose and cutting loads through data-feature-decision three-layer fusion, constructing a DQN and MPC hybrid self-adaptive planning model, optimizing with cutting efficiency, energy consumption and pick abrasion as multiple targets, and correcting paths in real time by combining closed loop feedback. Compared with the prior art, the coal rock identification accuracy rate reaches 98.7%, the pose positioning error is +/-8 mm, the cutting efficiency is improved by 18.3%, the energy consumption is reduced by 12.7%, the pick abrasion is reduced by 21.5%, the complex geological adaptation rate reaches 96.2%, the dust/water mist interference can be effectively inhibited, and the autonomous self-adaptive cutting under underground dynamic geology is realized.

Inventors

CHEN YAPING
LI YUNYUN
LUO YU
LEI FEI
HE MINGFEI
CHEN PEIHAO

Assignees

四川汇达幸福科技有限责任公司

Dates

Publication Date: 20260512
Application Date: 20260313

Claims (8)

1. The multi-sensor fusion self-adaptive cutting path planning method for the coal mine tunneling robot is characterized by comprising the following steps of: the method comprises the steps of firstly, disposing a laser radar, a millimeter wave radar, an IMU, a coal-rock spectrum sensor and a vibration sensor on a tunneling robot to form a five-source heterogeneous sensing system, and performing spatial registration with a machine body coordinate system through PTP high-precision time synchronization to complete multi-source data space-time unification; constructing a data-feature-decision three-layer fusion architecture, and sequentially carrying out denoising pretreatment, feature level fusion estimation and decision level reasoning output on multi-source perception information to obtain a coal-rock interface, a geological boundary, a robot pose and a cutting load state; Thirdly, establishing a dynamic geological model based on the fused environment and state information, and adopting a deep reinforcement learning DQN and model prediction control MPC hybrid algorithm to construct a multi-objective optimization function taking cutting efficiency, energy consumption and pick abrasion as cores so as to complete the on-line planning of a self-adaptive cutting path; And step four, a closed loop feedback correction mechanism is established, the planned path and the actual cutting state are compared in real time, and when the state deviation exceeds a threshold value, the path re-planning is triggered, so that the self-adaptive cutting control is realized.
2. The coal mine tunneling robot multi-sensor fusion self-adaptive cutting path planning method is characterized in that in the first step, time synchronization accuracy is better than +/-1 ms, the multi-sensor data after spatial registration are unified to a robot body Cartesian reference coordinate system, and the origin of the coordinate system is located at the center of gravity of the tunneling robot.
3. The multi-sensor fusion self-adaptive cutting path planning method of the coal mine tunneling robot is characterized in that in the second step, gaussian filtering and median filtering combined denoising algorithm is adopted in data layer fusion to finish noise reduction and time stamp alignment of point cloud, spectrum, vibration and gesture original data, improved self-adaptive Kalman filtering ACKF is adopted in feature layer fusion to combine Bayesian estimation to finish optimal estimation of multi-source feature information, D-S evidence theory is adopted in decision layer fusion, and decision results of coal rock types, geological contours, machine body pose and cutting load are output.
4. The coal mine tunneling robot multi-sensor fusion self-adaptive cutting path planning method is characterized by comprising the following steps of: wherein eta is cutting efficiency, P is energy consumption, W is pick wear amount, and the weighting coefficient satisfies + + =1, And =0.4、 =0.3、 =0.3。
5. The multi-sensor fusion self-adaptive cutting path planning method of a coal mine tunneling robot according to claim 4 is characterized in that in the third step, a DQN intelligent agent takes real-time geological state, robot pose and cutting load as input, a cutting head pose, cutting speed and cutting depth decision quantity are output, and MPC takes DQN output as a reference track to complete path optimization and constraint solving in a rolling time domain.
6. The multi-sensor fusion self-adaptive cutting path planning method of the coal mine tunneling robot is characterized in that in the fourth step, the triggering threshold value of closed loop feedback correction is that the robot pose deviation is larger than 10mm, the coal rock identification confidence is lower than 95%, the cutting load fluctuation range is larger than 20%, and the system completes path re-planning and instruction updating within 0.2s after triggering.
7. A multi-sensor fusion self-adaptive cutting path planning system of a coal mine tunneling robot is characterized by comprising a five-source heterogeneous sensing module, a space-time registration module, a three-layer fusion processing module, a DQN and MPC hybrid planning module and a closed-loop feedback control module, and is used for executing the multi-sensor fusion self-adaptive cutting path planning method of the coal mine tunneling robot according to any one of claims 1-6.
8. The multi-sensor fusion self-adaptive cutting path planning system of the coal mine tunneling robot is characterized by being mounted on a cantilever tunneling robot or a tunneling and anchoring integrated machine and used for realizing unmanned self-adaptive cutting operation in a complex underground environment with high dust, high water mist, faults and broken zones.

Description

Multi-sensor fusion self-adaptive cutting path planning method for coal mine tunneling robot Technical Field The invention relates to the technical field of intelligent tunneling of coal mines, in particular to a multi-sensor fusion self-adaptive cutting path planning method of a tunneling robot of a coal mine. Background Coal mine tunneling is a core link of coal mining, and a tunneling robot is core equipment for realizing intelligent and unmanned operation. Cutting path planning directly determines tunneling efficiency, forming quality, equipment life and operation safety. The prior art has the following defects: 1. The sensing is single, the anti-interference performance is poor, namely, the combination of a laser radar and an IMU is adopted, the point cloud is invalid under dust/water mist, the positioning drift is realized, the coal and rock identification depends on vision/vibration single source, the accuracy is low, and the single-layer Kalman filtering is adopted, the layering coordination is avoided, the error accumulation is large, and the pose error is large; 2. the path planning is not self-adaptive, mainly comprises offline pre-planning, cannot adapt to dynamic geology such as faults, broken bands and the like, has the dynamic response time exceeding 1s, only pursues the cutting efficiency, ignores the energy consumption and the cutting pick abrasion, and has high equipment failure rate and short service life. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a multi-sensor fusion self-adaptive cutting path planning method for a coal mine tunneling robot, which solves the problems of low sensing precision, inaccurate coal and rock identification, non-self-adaptive path, multi-target unbalance, closed loop deletion and the like in the background art. The multi-sensor fusion self-adaptive cutting path planning method for the coal mine tunneling robot comprises the following steps of: the method comprises the steps of firstly, disposing a laser radar, a millimeter wave radar, an IMU, a coal-rock spectrum sensor and a vibration sensor on a tunneling robot to form a five-source heterogeneous sensing system, and performing spatial registration with a machine body coordinate system through PTP high-precision time synchronization to complete multi-source data space-time unification; constructing a data-feature-decision three-layer fusion architecture, and sequentially carrying out denoising pretreatment, feature level fusion estimation and decision level reasoning output on multi-source perception information to obtain a coal-rock interface, a geological boundary, a robot pose and a cutting load state; Thirdly, establishing a dynamic geological model based on the fused environment and state information, and adopting a deep reinforcement learning DQN and model prediction control MPC hybrid algorithm to construct a multi-objective optimization function taking cutting efficiency, energy consumption and pick abrasion as cores so as to complete the on-line planning of a self-adaptive cutting path; And step four, a closed loop feedback correction mechanism is established, the planned path and the actual cutting state are compared in real time, and when the state deviation exceeds a threshold value, the path re-planning is triggered, so that the self-adaptive cutting control is realized. Preferably, in the first step, the time synchronization precision is better than +/-1 ms, the multi-sensor data after spatial registration are unified to a Cartesian reference coordinate system of a robot body, and the origin of the coordinate system is located at the gravity center position of the tunneling robot. In the second step, the data layer fusion adopts a Gaussian filtering and median filtering combined denoising algorithm to finish the noise reduction and time stamp alignment of point cloud, spectrum, vibration and gesture original data, the feature layer fusion adopts improved self-adaptive Kalman filtering ACKF to combine with Bayesian estimation to finish the optimal estimation of multi-source feature information, and the decision layer fusion adopts a D-S evidence theory to output the decision result of coal rock type, geological profile, body pose and cutting load. Preferably, in the third step, the multi-objective optimization function expression is: wherein eta is cutting efficiency, P is energy consumption, W is pick wear amount, and the weighting coefficient satisfies ++=1, And=0.4、=0.3、=0.3。 Preferably, in the third step, the DQN agent takes real-time geological state, robot pose and cutting load as input, outputs cutting head pose, cutting speed and cutting depth decision quantity, and the MPC takes DQN output as reference track to complete path optimization and constraint solving in the rolling time domain. Preferably, in the fourth step, the triggering threshold value of closed loop feedback correction is that the pose deviation of the robot is larger than 10mm, t