Deep mine hidden water disaster hazard source detection robot and working method thereof

Abstract

The invention relates to the field of mine disaster prevention and control, and particularly discloses a deep mine hidden water disaster hazard source detection robot and a working method thereof, and the deep mine hidden water disaster hazard source detection robot comprises a robot body, a multi-mode detection module, an intelligent control module, an autonomous advancing mechanism, an energy system and a communication module. The robot body structure is integrated with a high-precision geophysical sensor, a water quality analyzer, a microseism monitor and an AI data processing unit; the traveling mechanism adopts a crawler belt and wheel type composite design, and adapts to complex bottom plate terrains; the detection module realizes omnibearing analysis from a macroscopic geological structure to micro pore water flow through a multi-scale data fusion technology; and the intelligent control module generates a water disaster risk early warning in real time based on a deep learning algorithm. And the robot autonomously navigates to a target area, executes a detection task, and transmits data to a ground monitoring center through a wireless network. The problems of limited detection range and poor real-time performance in the prior art are solved, and the deep mine water disaster prevention and control capability is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of mine disaster prevention and control technology, specifically to a robot for detecting hidden water hazards in deep mines and its working method. Background Technology

[0002] Water inrush at the bottom of deep mines has become one of the core risks threatening mine safety. As mining depths extend below 800 meters, complex hydrogeological conditions lead to frequent water inrush accidents at the bottom. Statistics show that approximately 65% ​​of water inrush accidents in Chinese coal mines are related to confined water at the bottom. These water hazards are characterized by their suddenness and destructive power: on the one hand, deep rock strata are subjected to the coupled effects of high ground stress and high osmotic pressure, causing even tiny fissures to evolve into through-flow water channels within hours, triggering instantaneous water inrushes exceeding 1000 cubic meters per second. 3 / h of catastrophic accidents; on the other hand, the dynamic interaction between the hidden aquifer and the mining-induced damage zone makes it difficult to capture the early signs of water hazards, and traditional prevention and control methods such as advanced drilling and grouting reinforcement are seriously lagging behind.

[0003] Even more serious is the fact that deep mines often face multiple coupled threats from water hazards, including karst water, water accumulation in old workings, and fault-conducted water. The mechanisms underlying these hazards involve multiple disciplines, including geological structure, rock mechanics, and seepage dynamics, making it difficult for existing single monitoring methods to comprehensively assess the risks. According to industry reports, direct economic losses due to floor water hazards exceeded 12 billion yuan between 2018 and 2023, resulting in numerous serious casualties, highlighting the inadequacy of existing technological systems in the prevention and control of deep mine water hazards.

[0004] Currently, the main technologies for detecting water hazards in the aquifer rely on manual drilling and offline geophysical exploration. The former obtains local hydrological parameters through boreholes spaced 50-100m apart, but it has two major drawbacks: first, single-hole data only reflects information within a diameter of 2-3m, making it difficult to capture large-scale seepage anomalies; second, the drilling cycle is as long as 3-5 days, making dynamic monitoring impossible. The latter, while capable of regional scanning using resistivity and transient electromagnetic methods, has a spatial resolution limited to a detection accuracy of 10-20m and cannot distinguish between static aquifers and active water-conducting channels.

[0005] More importantly, existing methods generally suffer from the problem of data silos: geophysical data, water quality parameters and stress-strain monitoring data belong to independent systems, lacking a multi-source information fusion mechanism, resulting in a false alarm rate of 30%-40% for early warning models.

[0006] In recent years, although some studies have attempted to introduce robotic technology (such as the water quality testing robot described in CN 101234567A), these robots are only equipped with a single sensor and cannot simultaneously acquire geological structure data, nor do they possess autonomous obstacle avoidance or intelligent decision-making capabilities. For example, this patented robot needs to rely on a preset track to move, cannot adapt to undulating terrain, and data analysis relies entirely on manual interpretation, with a response delay of more than 2 hours.

[0007] Therefore, there is an urgent need to develop an intelligent detection equipment that integrates multi-scale detection, real-time data fusion, and autonomous operation to break through the technical bottleneck of water hazard prevention and control in deep mines. Summary of the Invention

[0008] To address the aforementioned problems in the prior art, this invention provides a robot for detecting hidden water hazards in deep mines and its working method, effectively solving the problems of poor real-time performance, low spatial resolution, and insufficient data utilization in the prior art, and effectively improving the identification accuracy of hidden water hazard sources.

[0009] To achieve the above objectives, this invention proposes a robot for detecting hidden water hazards in deep mines and its working method, comprising: a robot body, a multimodal detection module, an intelligent control module, an autonomous movement mechanism, an energy system, and a communication module. The robot body includes a pressure-resistant and explosion-proof shell, and its interior is divided into a detection compartment, a control compartment, and an energy compartment.

[0010] Preferably, the multimodal detection module integrates a high-resolution resistivity imager, a microseismic sensor array, a laser-induced breakdown spectroscopy (LIBS) instrument, a multi-parameter water quality sensor, a pore water pressure sensor, and a three-dimensional sonar imaging unit. The LIBS instrument is integrated within the detection chamber and acquires sample spectra through a sapphire glass optical window at the front of the chamber. The robot also includes a retractable robotic arm deployed on the outer front of the detection chamber. The pore water pressure sensor is embedded at the end of the robotic arm for directly measuring the pore water pressure in the rock strata. The three-dimensional sonar imaging unit is installed on the top of the detection chamber for constructing a three-dimensional structural model of the underlying strata.

[0011] Preferably, the intelligent control module has a built-in AI processor for real-time data fusion and risk modeling.

[0012] Preferably, the autonomous movement mechanism adopts a composite structure of tracks and multi-degree-of-freedom wheel sets, and is equipped with a terrain adaptive algorithm.

[0013] Preferably, the autonomous movement mechanism further includes a terrain recognition camera and a lidar, and a hydraulic lifting chassis. The terrain recognition camera and lidar are used to generate a movement path in real time; the hydraulic lifting chassis adjusts the ground clearance to adapt to uneven ground surfaces.

[0014] Preferably, the energy system includes a high-density lithium battery and a wireless charging interface.

[0015] Preferably, the communication module supports 5G / fiber hybrid transmission for real-time interaction with the ground monitoring center.

[0016] A working method for a robot for detecting hidden water hazards in deep mines, comprising the following specific steps for comprehensive detection of hidden water hazard sources in deep mines: S1. After receiving the detection task, the robot autonomously navigates to the target area based on the preset map and the real-time positioning system (RTK). S2. Activate the multimodal detection module to simultaneously collect resistivity, microseismic waves, water quality, and pore water pressure data; S3. Use an AI processor to fuse data, generate a probability map of water damage risk and mark potential hazards; S4. Transmit the early warning information and the three-dimensional geological model to the monitoring center to trigger the emergency response mechanism.

[0017] Therefore, this invention proposes a robot for detecting hidden water hazards in deep mines and its working method, the beneficial effects of which are as follows: Breaking through the spatial and temporal limitations of traditional offline detection methods, it achieves dynamic and continuous monitoring of the entire working face, eliminates blind spots in manual inspections, integrates multi-scale physical-chemical-mechanical sensor data, establishes a correlation model between geological structure, seepage field, and rock mass damage, improves the accuracy of identifying hidden water hazard sources, constructs an intelligent early warning system, and shortens the response time from data acquisition to risk assessment from hours to seconds, significantly enhancing the prevention and control capabilities of water inrush accidents. It effectively solves the core problems of poor real-time performance, low spatial resolution, and insufficient data utilization in existing technologies, providing all-weather, multi-dimensional water hazard prevention and control guarantees for safe production in deep mines.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall structure of a robot for detecting hidden water hazards in deep mines and its working method, according to the present invention. Figure 2 This is a schematic diagram of the multimodal detection module composition of a robot for detecting hidden water hazards in deep mines and its working method, as described in this invention. Figure 3 This is a flowchart of the autonomous navigation algorithm for a robot for detecting hidden water hazards in deep mines and its working method, according to the present invention. Figure 4 This is a data fusion and early warning system architecture diagram of a robot for detecting hidden water hazards in deep mines and its working method, according to the present invention.

[0020] Figure Labels 1. Pressure-resistant and explosion-proof outer shell; 2. Detection chamber; 3. Control chamber; 4. Energy chamber; 5. Telescopic robotic arm; 6. Sapphire glass optical window; 7. Three-dimensional sonar imaging unit. Detailed Implementation

[0021] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below. The described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of this application.

[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0023] This invention provides a robot for detecting hidden water hazards in deep mines and its working method, comprising: The robot body consists of a multimodal detection module, an intelligent control module, an autonomous movement mechanism, an energy system, and a communication module. The robot body includes a pressure-resistant and explosion-proof shell, and its interior is divided into a detection compartment, a control compartment, and an energy compartment.

[0024] The multimodal detection module integrates a high-resolution resistivity imager, a microseismic sensor array, a laser-induced breakdown spectrometer (LIBS), a multi-parameter water quality sensor, a pore water pressure sensor, and a three-dimensional sonar imaging unit. The LIBS is integrated inside the detection chamber and collects sample spectra through a sapphire glass optical window at the front of the chamber.

[0025] The robot also includes a retractable robotic arm deployed on the outer front of the probe cabin, with a pore water pressure sensor embedded at the end of the robotic arm for directly measuring the pore water pressure in the rock strata; and a three-dimensional sonar imaging unit for constructing a three-dimensional structural model of the underlying strata.

[0026] The intelligent control module has a built-in AI processor for real-time data fusion and risk modeling.

[0027] The autonomous movement mechanism employs a composite structure of tracks and multi-degree-of-freedom wheels, equipped with a terrain-adaptive algorithm. It also includes terrain recognition cameras and LiDAR, as well as a hydraulically lifted chassis. The terrain recognition cameras and LiDAR are used to generate the movement path in real time; the hydraulically lifted chassis adjusts its ground clearance to adapt to uneven terrain.

[0028] The energy system includes high-density lithium batteries and a wireless charging interface.

[0029] The communication module supports 5G / fiber hybrid transmission for real-time interaction with the ground monitoring center.

[0030] The specific steps for using a deep mine hidden water hazard detection robot to conduct comprehensive detection of hidden water hazard sources in deep mines are as follows: S1. After receiving the detection task, the robot autonomously navigates to the target area based on the preset map and the real-time positioning system (RTK). S2. Activate the multimodal detection module to simultaneously collect resistivity, microseismic waves, water quality, and pore water pressure data; S3. Use an AI processor to fuse data, generate a probability map of water damage risk and mark potential hazards; S4. Transmit the early warning information and the three-dimensional geological model to the monitoring center to trigger the emergency response mechanism.

[0031] The application of a robot for detecting hidden water hazards in deep mines and a method for using the robot for comprehensive detection of hidden water hazards in deep mines is as follows: The pressure-resistant and explosion-proof body adopts a compartmentalized titanium alloy structure with a compressive strength of not less than 200MPa. It integrates detection, control and energy subsystems and is suitable for high temperature and high humidity environments underground.

[0032] The multi-modal detection module simultaneously operates high-density resistivity imaging with a spatial resolution of 0.5m, achieving high positioning accuracy. Microseismic wave monitoring at 1m and LIBS spectral analysis with an elemental detection limit of 0.1ppm enable cross-scale diagnosis from macroscopic water-conducting channels to microscopic fissure seepage.

[0033] The intelligent control center is based on a U-Net+LSTM hybrid model, which performs feature-level fusion of multi-source data, dynamically generates a risk probability cloud map with an update frequency of 1Hz, and combines it with historical data to predict trends.

[0034] The autonomous driving system achieves path planning with zero human intervention in terrain with a slope angle of ≤35° by using a 20Hz scanning frequency three-dimensional lidar and adaptive hydraulic suspension in coordinated control.

[0035] The energy and communication system employs wireless charging with a charging efficiency of ≥85% and 5G edge computing technology to ensure 48 hours of continuous operation and real-time data transmission. The modules are deeply coupled through a bus architecture, forming a closed-loop control system of detection, analysis, and early warning.

[0036] like Figures 1-2As shown, the robot body is encased in a titanium alloy explosion-proof shell 1, which is internally divided into a detection cabin 2, a control cabin 3, and an energy cabin 4. A retractable robotic arm 5 is deployed on the outer front of the detection cabin, and a sapphire glass optical window 6 of LIBS is provided at the front of the detection cabin. A three-dimensional sonar imaging unit 7 is installed on the top.

[0037] The bottom of the probe is equipped with a resistivity imaging electrode array, and the robotic arm can extend to the surface of the rock wall to directly measure water pressure data through the pore water pressure sensor at the end.

[0038] (1) Robot pressure-resistant and explosion-proof body: The robot's pressure-resistant and explosion-proof body adopts a compartmentalized streamlined structure, with an overall hexagonal prism shape (length × width × height = 1.8m × 1.2m × 0.9m). The main body is made of TC4 titanium alloy (tensile strength ≥ 950MPa, yield strength ≥ 880MPa), with a nitrided surface forming a dense oxide layer of 20-30μm, capable of withstanding downhole instantaneous impact pressure of 1.5MPa and corrosive environments with pH 2-12. The outer shell features a double-sealed structure: a 5mm thick stamped titanium alloy outer shell and a 3mm fluororubber gasket inner layer. The seams are sealed using laser welding and O-rings, achieving an IP68 protection rating (no leakage after continuous immersion in 2m water for 72 hours).

[0039] The key sensor windows are covered with sapphire glass (transmittance ≥92%), and a dynamic pressure balancing valve enables adaptive adjustment of internal and external air pressure, ensuring no condensation or deformation within a temperature range of -30℃ to 85℃. Explosion-proof certification complies with GB3836.1-2010 standards, allowing safe operation in explosive environments with methane concentrations ≤15%.

[0040] The robot's internal structure adopts a "three-compartment separation" architecture: the detection compartment is located at the front of the robot, integrating a high-density resistivity imaging electrode array (64 channels), microseismic sensors (8 sets of triaxial detectors) and a telescopic robotic arm (1.2m in unfolded length). The compartment wall is embedded with an electromagnetic shielding layer (shielding effectiveness ≥60dB) to ensure the accuracy of high-frequency signal acquisition. The control cabin is centrally located and houses a main control board (based on ARM+FPGA architecture), an AI processor (computing power ≥10TOPS), and a data storage unit (capacity 4TB), which is interconnected with each module at high speed via PCIe bus; The energy compartment is located at the stern and is equipped with a 48V / 120Ah lithium-sulfur battery pack and a wireless charging receiver coil (power transmission efficiency ≥85%). The compartment and the outer shell are filled with an aluminum silicate ceramic fiber insulation layer (thermal conductivity ≤0.03W / m·K).

[0041] Each compartment is connected via quick-release flanges, allowing for module replacement within 15 minutes. For structural reinforcement, the internal structure utilizes honeycomb titanium alloy reinforcing ribs (8mm thick, 50mm spacing), achieving an overall compressive strength ≥200MPa and capable of withstanding static loads from 100m of rock. The cooling system consists of heat pipes (heat transfer power ≥200W) and heat dissipation fins (surface area 2.4m²). 2 Composed of [missing information - likely components], and equipped with dual turbine fans (40 CFM airflow), ensuring that the temperature rise of internal components is ≤25℃. Explosion-proof connectors adopt the MIL-DTL-38999 standard to achieve reliable signal and power transmission between modules.

[0042] (2) Multimodal detection module: The multimodal detection module employs a three-dimensional sensing system integrating physics, chemistry, and mechanics to achieve comprehensive diagnosis of water hazard sources in deep mine floors. The module's core functions include: Geological structure tomography: A three-dimensional electrical structure model of the base plate within a depth of 50m was constructed using the high-density resistivity method (electrode spacing 0.5m, maximum power supply current 5A) to identify macroscopic water-conducting channels such as karst development zones and fault fracture zones in the base plate (spatial resolution ≤0.5m). Dynamic seepage field monitoring: Deploy 8 sets of three-component microseismic sensors (frequency band 0.1-500Hz, sensitivity 100V / m / s) to capture rock mass fracture events in real time (positioning accuracy). 1m), combined with acoustic emission energy counting (threshold > 50dB) to invert the fracture propagation rate; Hydrochemical characterization analysis: Laser-induced breakdown spectroscopy (LIBS) was employed. The analysis unit is an integrated module consisting of a laser emitter, a spectral acquisition unit, and a signal processor, used for rapid in-situ elemental analysis. This unit utilizes dual-pulse laser excitation (50° interval). In-situ elemental analysis (Ca) was performed on the water seepage from the cracks. 2+ Mg 2+ Plasma spectral line resolution ≤ 0.1 ), synchronously integrating a multi-parameter water quality sensor (pH) 0.1, conductivity 1 ( / cm, dissolved oxygen ±0.2mg / L), establish a water chemical fingerprint database to trace water sources; Direct measurement of pore water pressure: A miniature pore water pressure gauge (range 0-20MPa, accuracy 0.5%FS) is embedded into the rock surface through a telescopic robotic arm (expansion accuracy ±1mm) to obtain transient water pressure fluctuation data in millimeter-level fissures; Multi-source data fusion diagnosis: Based on feature-level data association algorithms, resistivity anomaly areas (>100Ω·m) and microseismic event clusters (spatial density >5 events / m) are integrated. 3Spatiotemporal matching was performed between the water hazard risk probability model and abrupt changes in water chemistry indicators (Ca / Mg ratio change rate > 20%) to construct a water hazard risk probability model (confidence level ≥ 90%).

[0043] This model is built on a Bayesian network. It takes the aforementioned three dimensions of feature parameters as input and trains on a historical flood inrush case dataset (containing 3000+ samples) to determine the weights of each parameter (resistivity anomaly weight). =0.4, weight of microseismic event clusters =0.35, weight of abrupt changes in water chemical indicators =0.25), the formula for quantifying risk probability is: ; In the formula, This represents the risk probability corresponding to resistivity anomalies, ranging from 0 to 1, and is obtained by mapping the degree of deviation between the resistivity value and the threshold. The risk probability corresponding to the microseismic event cluster, with a value range of 0-1, is obtained by mapping the degree of deviation between the event density and the threshold. The risk corresponding to abrupt changes in water chemical indicators is represented by a probability value ranging from 0 to 1. It is obtained by mapping the deviation between the ratio change rate and the threshold, and the confidence level is ensured to be ≥90% through 5-fold cross-validation.

[0044] The multimodal detection module also scans the surface topography of the base plate using three-dimensional sonar (frequency 200kHz) to generate a digital elevation model (DEM) with millimeter-level accuracy, providing a topographic benchmark for seepage path analysis.

[0045] The specific technical implementation of the multimodal detection module includes: First, the high-density resistivity imaging system consists of a ring array of 64 titanium alloy electrodes (1.2m in diameter), using a Winner-Schlumberger combination device (AB=MN=0.5m). Full-section scanning is achieved through a multi-channel switcher (switching time <10ms). The data acquisition card (24-bit ADC, sampling rate 1kS / s) converts the potential difference signal into a resistivity distribution map. Combined with an improved finite element inversion algorithm (regularization parameter λ=0.01), real-time three-dimensional imaging is completed on the FPGA chip (single frame processing time <2s). Secondly, the microseismic monitoring network contains eight MEMS accelerometers (noise density 50 μg / √Hz), arranged in an icosahedral pattern on the inner wall of the detection chamber. It extracts valid events through an adaptive threshold triggering mechanism (STA / LTA=3 / 10) and uses a layered velocity model (P-wave velocity 4500 m / s, S-wave velocity 2800 m / s) and Geiger localization algorithm to achieve precise location of the rupture source. Next, the LIBS spectral analysis unit uses dual-pulse laser excitation (50ns interval), and the spectra are dispersed by a Czerny-Turner spectrometer (500mm focal length, 2400g / mm grating). An EMCCD detector (quantum efficiency >90%, -70℃ cooling) captures the 350-900nm band spectrum, and the ion concentration is quantitatively inverted by combining the partial least squares regression (PLSR) algorithm. Furthermore, the robotic arm measurement system integrates a six-degree-of-freedom collaborative robotic arm (with repeatability accuracy of ±0.02mm), and is equipped with a miniature hydraulic pressure gauge (8mm in diameter) and a high-definition camera (5 megapixels) at the end. It achieves crack location and contact force feedback (threshold 0.5N) through visual servo control, and the measurement data is transmitted to the main control unit via CAN bus. Finally, the data fusion hub runs a U-Net-based semantic segmentation model (input layer 512×512 resistivity image) and an LSTM temporal prediction network (time window 60s), achieving feature-level fusion of multimodal data on the NVIDIA Jetson AGX Xavier platform (frame synchronization error <10ms), and outputting a JSON format early warning message containing risk level (low / medium / high), confidence level and spatial coordinates. The quantitative standard for risk level is: risk probability. It was determined to be low risk at that time. It was determined to be of medium risk at the time. It was determined to be high risk at that time.

[0046] The various subsystems within the multimodal detection module are synchronized via a time synchronization module (GPS / BeiDou dual-mode, synchronization accuracy). Achieving spatiotemporal alignment of data (1μs) ensures consistency in cross-scale diagnostics.

[0047] (3) Functional architecture of intelligent control center: The intelligent control center constitutes the decision-making core of the robot, possessing four core functions: Multi-source heterogeneous data fusion: Through feature-level and decision-level fusion mechanisms, resistivity imaging data (spatial grid 0.5m×0.5m) and microseismic event coordinates (…) are fused together. Spatiotemporal alignment of LIBS spectral data (0.1ppm detection limit) and pore water pressure time-series signal (sampling rate 100Hz) was performed to construct a multi-dimensional feature matrix; Dynamic risk assessment modeling: Geological anomaly areas are defined as areas that meet any of the following conditions: resistivity below 50 Ω·m or above 100 Ω·m; spatial density of microseismic event clusters > 5 events / m. 3 Sudden changes in water chemistry parameters (Ca / Mg ratio change rate > 20%). Semantic features of the above-mentioned geological anomaly areas were extracted based on the U-Net convolutional neural network (input layer 512×512 pixels). The temporal evolution of seepage pressure was analyzed by combining the LSTM network (hidden layer 256 nodes). A risk probability heat map (spatial resolution 0.1m, update frequency 1Hz) was generated in real time using the risk probability formula. Adaptive early warning optimization: The Bayesian dynamic threshold algorithm is adopted to automatically adjust the early warning triggering conditions based on the historical water inrush case library (containing 3000+ samples) and real-time environmental parameters (temperature, humidity, methane concentration) (false alarm rate <5%). System coordinated control: The sampling frequency of the detection module (resistivity scanning period of 10s, continuous acquisition of microseismic data), motion parameters of the traveling mechanism (speed adjustable from 0-1.5m / s), and data transmission priority of the communication module (5G bandwidth allocation strategy) are scheduled through a CAN / Ethernet hybrid bus to ensure optimal resource allocation for the entire system.

[0048] In addition, the central integrated autonomous learning engine continuously optimizes the detection strategy and path planning logic through reinforcement learning algorithms (Q-learning, the reward function of which includes risk identification accuracy and energy efficiency indicators).

[0049] The central hardware adopts a heterogeneous computing architecture: the main control unit is equipped with a Xilinx Zynq UltraScale+ MPSoC (quad-core ARM Cortex-A53 + FPGA), which is responsible for sensor data preprocessing (finite difference inversion of resistivity data); the AI ​​acceleration module uses NVIDIA Jetson AGX Xavier (32 TOPS computing power), deploying a U-Net (5 layers deep encoder, 3×3 convolution kernels) + LSTM (60 time steps) hybrid model. The training dataset contains 50,000 sets of labeled water inrush precursor data, with a training accuracy of 98.7%.

[0050] At the software level, a three-layer data pipeline is established: the real-time processing layer uses ZeroMQ to implement multi-threaded data reception (throughput ≥1GB / s) and adopts a sliding window mechanism (window size 60s, step size 1s) for feature extraction; The analysis of the decision-making layer running the TensorRT-optimized neural network model showed that the processing time for a single frame resistivity image was <50ms, and the LSTM network predicted the seepage pressure change in the next 30 minutes (RMSE <0.15MPa). The control output layer issues control commands through the ROS2 framework. The risk heat map is encapsulated in GeoJSON format (containing three-dimensional coordinates of longitude, latitude, and depth, and risk values ​​of 0-1). After the warning signal is triggered, the sound and light alarm (105dB buzzer + RGB LED strobe) is automatically activated and the encrypted data packet is uploaded to the ground server (AES-256 encryption, transmission delay <200ms).

[0051] The time synchronization system employs the PTP precision clock protocol (master-slave clock deviation <1μs) to ensure timescale consistency of multi-sensor data. The data caching module is equipped with 8GB DDR4 memory and a 1TB NVMe solid-state drive, enabling continuous storage of 72 hours of raw data even during communication interruptions.

[0052] (4) Autonomous Mobility System: The autonomous mobility system enables the robot to adapt to all terrains and achieve unmanned exploration operations in complex downhole environments.

[0053] The core functions of the system include: 3D environment perception: A 32-line LiDAR (scanning frequency 20Hz, angular resolution 0.1°) and a binocular vision camera (global shutter, frame rate 30fps) generate a point cloud map with centimeter-level accuracy (density ≥16000 points / ㎡) in real time to identify unevenness of the base plate (height difference ≥50mm), water accumulation areas (depth >100mm) and obstacles (size ≥200mm). Dynamic Path Planning: Based on Improved RRT The algorithm (convergence time < 500ms) and the prior geological model are used to plan safe routes (minimum turning radius 0.6m) in sloping terrain with a dip angle ≤ 35°, and the microseismic risk heat map is fused simultaneously (areas with probability > 0.7 are set as no-entry zones). Multimodal motion control: It adopts a six-wheel independent drive architecture (300mm wheel diameter, rubber tread), combined with a hydraulic suspension system (150mm travel) and differential lock mechanism to achieve stable driving with an obstacle crossing height of ≥300mm and a side roll angle compensation of ±25°; Autonomous obstacle avoidance and recovery: When a sudden obstacle (such as falling rocks) is detected, the trajectory is adjusted in real time through the DWA local path optimization algorithm (update frequency 10Hz). If it gets stuck in the mud area (wheel slip rate > 40%), the track assist mode is activated (deployment speed < 3s). Energy optimization management: Terrain complexity is a parameter that comprehensively reflects the undulation of the terrain and the difficulty of passage. The slope is calculated using point cloud data collected by lidar, and the calculation formula is as follows: Slope = arctan(elevation difference / horizontal distance); The coefficient of friction is estimated using feedback data from wheel speed and torque sensors, with a real-time acquisition frequency of 10Hz. The formula for calculating the coefficient of friction is as follows: Coefficient of friction = torque / (wheel radius × vehicle weight × adhesion coefficient); The system dynamically adjusts the motor torque output (0-200Nm continuously variable transmission) based on terrain complexity (slope, coefficient of friction), and the power consumption fluctuation range is controlled within a certain range. With a margin of error of less than 15%, the system can guarantee continuous operation for 48 hours. It can also receive remote commands via a 5G network, enabling seamless switching between semi-autonomous and fully autonomous modes.

[0054] (5) Energy and communication systems: The energy and communication system forms the core support system for robots' persistent operation and real-time interaction, and includes the following core functions: High-efficiency energy supply: Modular lithium-sulfur battery packs (energy density ≥400Wh / kg) are used to provide differentiated power distribution to the detection module (peak power 1200W), the traveling mechanism (continuous power 800W) and the computing unit (dynamic power consumption 50-200W) through intelligent time-sharing power supply strategy, ensuring 48 hours of continuous operation; Wireless charging and energy self-sufficiency: Based on magnetic resonance coupling technology (operating frequency 85kHz), non-contact power transmission (efficiency ≥85%) is achieved within 3m of the charging base station, which, together with the photovoltaic auxiliary unit (energy conversion efficiency of 18% for underground emergency lighting), forms a multi-source power supply network. High-reliability communication relay: Integrates 5G millimeter wave (bandwidth 400MHz, rate 10Gbps) and mining fiber (single-mode, loss ≤0.3dB / km) dual-mode communication, and achieves seamless switching (switching delay <50ms) through dynamic link quality assessment algorithm (RSSI + bit error rate joint criterion), ensuring real-time transmission of multimodal detection data (daily average data volume ≥1TB) and risk warning signals; Intelligent energy management: Built-in multi-parameter monitoring chip (sampling rate 1kHz) tracks battery health status in real time (SOH accuracy). 2%), temperature gradient ( The charge / discharge curve was optimized using a reinforcement learning algorithm, taking into account the temperature (0.5℃) and the number of charge / discharge cycles (resulting in a 30% improvement in cycle life). Data security and redundancy: The system adopts the national cryptographic SM4 / AES-256 dual encryption protocol to protect the transmitted data end-to-end, and establishes a dual storage mechanism of local cache (capacity 4TB) + cloud synchronization. In the event of communication interruption, the system automatically activates the LoRa self-organizing network (transmission distance 2km) to relay data.

[0055] The system technology implementation includes: The perception unit consists of a Velodyne VLP-32C LiDAR (detection range 100m, accuracy ±3cm) and a FLIRBlackfly S binocular camera (IMX535 sensor, pixel size 3.45μm). A real-time 3D semantic map (grid size 0.1m×0.1m) is constructed through a GPU-accelerated point cloud registration algorithm (ICP variant, iteration count ≤50). The decision-making unit is equipped with an NVIDIA Orin processor (275 TOPS computing power) and runs a hierarchical path planning algorithm—the global layer uses A The algorithm combines geological structural data (fault strike, aquifer distribution), and uses the TEB optimizer (time elastic band) for dynamic obstacle avoidance in local layers, with a planning delay of <200ms; The actuator uses six brushless motors (peak power 800W) to drive the MacPherson independent suspension. The hydraulic system includes four bidirectional actuators (pressure 35MPa), and a PID controller (bandwidth 100Hz) enables millimeter-level adjustment of the suspension travel (position error). 2mm); The collaborative control unit integrates an IMU (zero-bias stability 0.1° / h), wheel speed encoder (resolution 0.01m / s), and suspension displacement sensor via a CAN FD bus (5Mbps transmission rate). It employs model predictive control (MPC, prediction time 2s) to achieve active vehicle attitude balancing (roll angle fluctuation <3°). For handling special scenarios: on a 30° slippery slope, the system automatically switches to diagonal wheel torque coupling mode (traction increased by 40%); when encountering a 20cm high step, the hydraulic suspension raises the chassis ground clearance to 250mm, combined with a swing-arm obstacle-crossing strategy (joint angle 0-90°) to achieve traversal. The fault recovery mechanism adopts a dual-redundancy design. When the main control fails, the backup PLC (response time <10ms) immediately takes over control and initiates an emergency return procedure.

[0056] The energy unit consists of 48V / 200Ah lithium-sulfur battery modules (50Ah capacity per cell, ≥2000 cycles with 80% DOD). It employs an active balancing BMS (10A balancing current) to manage 24 cell clusters. The heat dissipation system integrates heat pipes (thermal conductivity 398W / m·K) and liquid cooling plates (flow rate 2L / min) to maintain a temperature difference of ±5℃ in ambient temperatures ranging from -20℃ to 60℃. The wireless charging system employs a DDQ coil structure (500mm diameter transmitter, 300mm diameter receiver), achieving 85kHz resonance based on a SiC MOSFET inverter (1MHz switching frequency). It utilizes an adaptive impedance matching network (VSWR < 1.5) and PID power control (precision...) 5W), stably transmitting 3kW of power within an air gap range of 3cm-1m; The communication unit hardware layer deploys Huawei MH5000 5G industrial module (supporting NSA / SA dual mode) and gigabit fiber transceiver (wavelength 1310nm), while the software layer runs a customized TCP / IP protocol stack (RTO dynamic adjustment algorithm), which automatically switches to fiber optic communication when the 5G signal attenuates to -110dBm (switching time <30ms). The energy management hub is equipped with a TI BQ40Z80 chipset, which calculates the remaining battery life in real time (with an error of ±5 minutes) and sends energy-saving commands (such as reducing the LiDAR scanning frequency to 10Hz) to the control hub via the CAN bus. The data security module integrates an encrypted SOC (National Cryptographic Level 2 Authentication), employing a layered encryption strategy—metadata uses the SM4 algorithm (128-bit block size), raw probe data uses AES-256-GCM mode (96-bit authentication tag size), and the key is updated every 30 minutes via a quantum random number generator. The system also features a dual-power redundant design; in the event of a main battery failure, a backup supercapacitor bank (100F capacity) can support a 10-minute emergency return.

[0057] like Figure 3 As shown, the robot uses an improved AI algorithm to plan its path and constructs a point cloud map in real time using LiDAR. When an obstacle is detected, the hydraulic lifting chassis automatically rises to ensure smooth obstacle crossing.

[0058] like Figure 4 As shown, after receiving data from multiple sources, the AI ​​processor extracts features through a convolutional neural network and outputs the flood risk level. When the risk value exceeds the threshold, an audible and visual alarm is triggered, and the warning coordinates are uploaded to the monitoring center.

[0059] Applying a robot for detecting hidden water hazards in deep mines to detect water in the bottom slab: Before performing the task of detecting water hazards in the well floor, the robot first receives task instructions from the ground monitoring center via 5G communication and retrieves a pre-set 3D map and historical geological data from the well. Utilizing the built-in RTK high-precision positioning system and the point cloud map generated by lidar, combined with the geological structure model and mining progress information, the control center completes preliminary path planning based on an improved AI algorithm. Simultaneously, the system sets high-risk areas marked on the risk heat map as key scanning targets and excludes high-angle, water-filled, or obstacle-prone areas as prohibited paths, providing reliable navigation data for the efficient execution of subsequent tasks.

[0060] The robot activates its onboard six-wheel independent drive system, combining a hydraulic lifting chassis and differential control technology to enter autonomous driving mode. During actual operation, it acquires micro-topographical information of the chassis surface using a 20Hz frequency 3D LiDAR and binocular cameras, identifying obstacles such as pits, puddles, and debris in real time. If it encounters obstacles with a height difference ≥20cm or a slope angle exceeding 25°, the system activates the hydraulic suspension to adjust ground clearance and deploys track-assisted driving mode to ensure obstacle-crossing stability. Based on the TEB optimizer, the system updates the path every 0.1 seconds, enabling the robot to adapt to the complex and varied terrain features of the chassis and ensuring uninterrupted continuous operation.

[0061] Once the robot reaches the designated base area, it automatically deploys the bottom electrode array and the resistivity imaging module. The system controller sequentially activates the electrode pairs, using a Winner-Schlumberger combination device for current injection and potential difference acquisition. The acquired data is converted from analog to digital by a 24-bit ADC and processed in an FPGA chip using an improved finite element inversion algorithm to quickly construct a three-dimensional electrical model within a 50m depth. The imaging results are output with a spatial resolution of 0.5m, clearly identifying low resistivity anomaly zones, predicting potential karst water-conducting structures, fracture zones, and the spatial distribution of aquifers, and providing preliminary information on the location of water hazard sources.

[0062] To identify active fractures and water inrush channels within the bedrock, eight MEMS microseismic sensors within the robotic detection cabin operated continuously, monitoring a frequency range of 0.1–500 Hz. The system employed a STA / LTA event-triggered mechanism to extract valid waveforms and performed three-dimensional source inversion based on a layered velocity model (P-wave / S-wave) and the Geiger localization algorithm, achieving an accuracy of ±1 m. Energy counting and source density analysis determined the presence of concentrated fracture zones or fracture propagation trends, providing crucial evidence for understanding the dynamics of hazardous water bodies and potential water inrush mechanisms.

[0063] After locating a suspected leakage fissure, the robotic arm automatically extends and uses a high-definition vision system to pinpoint the target. A microporous water pressure sensor on its end effector inserts into the fissure surface, recording water pressure fluctuations in real time and identifying the level of seepage activity. Simultaneously, a LIBS laser-induced spectrometer performs elemental analysis on the seepage water, identifying calcium... 2+ Mg 2+ Fe 2+ Changes in the content of components, combined with parameters such as pH and conductivity, are used to determine the type of water source and its correlation with old abandoned areas, fault water, or karst water. The specific determination method is as follows: Establish a database of hydrochemical characteristics of water from old goaf areas, fault water, and karst water (water from old goaf areas: Ca 2 + content <50mg / L, pH <7; Fault water: Mg 2 + content >80mg / L, conductivity >500μS / cm; karst water: Fe2 +Content>30mg / L), the measured data are matched with the database using cosine similarity. If the similarity is ≥85%, it is determined to be the corresponding water source type. The strength of the correlation is quantified by the similarity value (85%-90% is weak correlation, 90%-95% is moderate correlation, and above 95% is strong correlation), providing a quantitative basis for subsequent risk level classification.

[0064] Data collected by various sensors is transmitted to the main control center via a CAN bus and processed uniformly in the AI ​​processor. The U-Net semantic segmentation network extracts anomaly features from the resistivity image, while the LSTM model predicts trends in water pressure and seismic time-series signals. The system integrates multi-source information to generate a three-dimensional risk heat map. This map quantifies the risk level in the range of 0–1 and overlays downhole coordinates to achieve spatial visualization. When the risk value in a local area exceeds a set threshold, the system immediately triggers an audible and visual alarm and uploads the warning data to the ground center via the 5G network to assist decision-makers in emergency response.

[0065] During the exploration mission, if the robot detects drastic terrain changes (such as collapsed boundaries or surface anomalies caused by sudden water flow), it will provide real-time feedback via IMU, wheel speedometer, and lidar, and adjust its current path in conjunction with a risk heat map. The control center uses an MPC model prediction algorithm to predict attitude change trends within 2 seconds, thereby optimizing the obstacle avoidance trajectory. In case of operational anomalies, such as getting stuck in a mud pit or experiencing a communication interruption, the backup PLC controller will take over the system within 10ms, activating the emergency return-to-home strategy and recalling historical safe paths to terminate the mission and protect the equipment.

[0066] The robot generates ≥1TB of data stream daily throughout its operation cycle. All raw data is encrypted with AES-256-GCM and stored in layers on a local 1TB NVMe SSD and a cloud backup center. Simultaneously, risk reports and high-frequency heat maps generated in GeoJSON format are provided in a structured form for use by gas monitoring and roof pressure monitoring systems, enabling joint prevention and control of underground disasters. Furthermore, after the mission, the robot returns to the base station area, recharges via a wireless charging system, and simultaneously performs full data synchronization and AI model iterative training through a fiber optic interface, continuously improving future detection accuracy.

[0067] Therefore, this invention provides a robot for detecting hidden water hazards in deep mines and its working method, which effectively solves the core problems of poor real-time performance, low spatial resolution and insufficient data utilization in the prior art. It provides all-weather, multi-dimensional water hazard prevention and control for safe production in deep mines, has wider applicability, and can be deployed in deep mining faces such as coal mines and metal mines to achieve 24-hour continuous monitoring and intelligent early warning of floor water hazards, reduce the incidence of water inrush accidents, and ensure safe production in mines.

[0068] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A robot for detecting hidden water hazard sources in deep mines, characterized in that, include: The robot body comprises a multimodal detection module, an intelligent control module, an autonomous movement mechanism, an energy system, and a communication module. The robot body includes a pressure-resistant and explosion-proof shell, and its interior is divided into a detection compartment, a control compartment, and an energy compartment.

2. The robot for detecting hidden water hazards in deep mines according to claim 1, characterized in that, The multimodal detection module integrates a high-resolution resistivity imager, a microseismic sensor array, a laser-induced breakdown spectroscopy (LIBS) instrument, a multi-parameter water quality sensor, a pore water pressure sensor, and a three-dimensional sonar imaging unit. The LIBS instrument is integrated within the detection chamber and acquires sample spectra through a sapphire glass optical window at the front of the chamber. The robot also includes a retractable robotic arm deployed on the outer front of the detection chamber. The pore water pressure sensor is embedded at the end of the robotic arm for directly measuring the pore water pressure in the rock strata. The three-dimensional sonar imaging unit is installed on the top of the detection chamber for constructing a three-dimensional structural model of the underlying strata.

3. The robot for detecting hidden water hazards in deep mines according to claim 1, characterized in that, The intelligent control module has a built-in AI processor for real-time data fusion and risk modeling.

4. The robot for detecting hidden water hazards in deep mines according to claim 1, characterized in that, The autonomous movement mechanism adopts a composite structure of tracks and multi-degree-of-freedom wheel sets, and is equipped with a terrain adaptive algorithm.

5. The robot for detecting hidden water hazards in deep mines according to claim 1, characterized in that, The autonomous movement mechanism also includes a terrain recognition camera and a lidar, and a hydraulic lifting chassis. The terrain recognition camera and lidar are used to generate a movement path in real time; the hydraulic lifting chassis adjusts the ground clearance to adapt to uneven ground surfaces.

6. The robot for detecting hidden water hazards in deep mines and its working method according to claim 1, characterized in that, The energy system includes a high-density lithium battery and a wireless charging interface.

7. The robot for detecting hidden water hazards in deep mines according to claim 1, characterized in that, The communication module supports 5G / fiber hybrid transmission and is used for real-time interaction with the ground monitoring center.

8. A method for using a robot for detecting hidden water hazards in deep mines as described in any one of claims 1-7, characterized in that, The specific steps for conducting comprehensive detection of hidden water hazard sources in deep mines are as follows: S1. After receiving the detection task, the robot autonomously navigates to the target area based on the preset map and the real-time positioning system (RTK). S2. Activate the multimodal detection module to simultaneously collect resistivity, microseismic waves, water quality, and pore water pressure data; S3. Use an AI processor to fuse data, generate a probability map of water damage risk and mark potential hazards; S4. Transmit the early warning information and the three-dimensional geological model to the monitoring center to trigger the emergency response mechanism.

Images