RFID-based mining safety boot full-life-cycle traceability system

Abstract

The invention relates to the technical field of mine personnel positioning and safety production management, and discloses an RFID-based mining safety boot full life cycle traceability system, a sensing layer adopts an energy collection type semi-passive dual-mode tag deployed on a safety boot, the tag comprises a composite energy storage and collection unit and a dual-mode communication module, and the dual-mode communication module is fused with UWB and RFID; the processing layer comprises an underground edge node, an energy consumption optimization sub-module in a core collaborative algorithm is operated, vibration data are analyzed in real time through a lightweight one-dimensional expansion causal convolutional network, a roof weighting precursor is identified, and a UWB full function mode, an RFID ONLY mode or a deep sleep mode are dynamically switched in combination with a super capacitor voltage; and when a hidden danger occurs, the UWB is forcibly activated, the positioning frequency is improved, and unnecessary power consumption is inhibited. According to the method, the full-life-cycle closed-loop traceability of the safety boots from warehousing, service, maintenance to scrapping is realized, the operation and maintenance cost is remarkably reduced, and the underground emergency response capability is improved.

Description

Technical Field

[0001] This invention relates to the field of mine personnel positioning and safety production management technology, specifically to an RFID-based full lifecycle traceability system for mine safety boots. Background Technology

[0002] The underground coal mine working environment is complex, and personnel positioning and safety monitoring are core aspects of coal mine safety production. According to relevant standards such as AQ 6210-2007 "General Technical Conditions for Coal Mine Underground Worker Management System," underground workers must be equipped with positioning identification cards to achieve real-time location monitoring, attendance management, and emergency search and rescue. Currently, underground personnel positioning technology has mainly evolved from RFID (Radio Frequency Identification) to UWB (Ultra-Wideband): RFID technology utilizes the 920-925MHz frequency band for area identification and attendance management, featuring low cost and simple deployment; UWB technology, based on the IEEE 802.15.4a standard, utilizes the 6.5GHz frequency band to achieve centimeter-level high-precision positioning, meeting the precise positioning needs of complex underground roadways.

[0003] However, existing downhole positioning and identification technologies have the following technical shortcomings in practical applications:

[0004] (1) Lack of closed-loop management throughout the entire life cycle and serious waste of resources. The physical design life of existing mining safety boots is usually 3 years, while the built-in positioning tags (chips) are powered by button batteries and have a lifespan of only about 6 months. The current model adopts one-time binding, one-time use and overall scrapping, which means that 6 batches of tags need to be replaced within the 3-year life cycle of the safety boot (replaced every 6 months), resulting in high maintenance costs; or when the safety boot reaches the 3-year scrapping period, the tags with still good performance (used for only 6 months) are discarded along with the boot body, resulting in serious waste of electronic components.

[0005] (2) Lack of scientific assessment methods for component performance degradation under complex environments. Existing technologies mostly use button batteries for power supply, lacking online monitoring of battery remaining capacity and degradation status; at the same time, there is a lack of accelerated degradation assessment models for key performance parameters such as tag sealing and chip sensitivity under harsh underground environments (85℃ / 85%RH). This results in tags either being prematurely scrapped and wasted while their performance is still acceptable, or continuing to operate beyond their service life after performance degradation exceeds the safety threshold, posing serious safety hazards.

[0006] (3) Poor algorithm coordination and low level of intelligent control. The existing system's energy consumption management, timing scheduling, and encrypted transmission modules usually operate independently and lack a unified core coordination mechanism. For example, the energy consumption optimization strategy is not linked to the downhole safety status (such as vibration frequency shift caused by the precursor of roof pressure) and cannot dynamically adjust the positioning frequency according to the real-time working conditions; the data verification mechanism is imperfect and lacks the ability to cross-verify multimodal data, making it difficult to identify false data or label replacement; the cloud-edge coordination mechanism is weak and the cloud model optimization cannot be quickly synchronized to the downhole edge node, making it difficult to achieve real-time prediction of hidden dangers and self-adaptive optimization of algorithms. Summary of the Invention

[0007] This invention provides an RFID-based full lifecycle traceability system for mining safety boots. Through the deep integration of an energy-harvesting semi-passive dual-mode architecture and a core collaborative algorithm, it achieves closed-loop traceability of the safety boots from warehousing, service, maintenance to scrapping, while meeting the stringent constraints of intrinsic safety and explosion-proof requirements in underground coal mines. It balances high-precision positioning, intrinsic safety and explosion-proof features with intelligent energy efficiency management, significantly reducing operation and maintenance costs and improving underground emergency response capabilities.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] An RFID-based full lifecycle traceability system for mining safety boots includes a sensing layer, a transmission layer, a processing layer, and a core collaborative algorithm.

[0010] The sensing layer includes an energy-harvesting semi-passive dual-mode tag deployed on a mining safety boot. The tag includes a composite energy storage and harvesting unit and a dual-mode communication module. The composite energy storage and harvesting unit includes a piezoelectric ceramic array, a supercapacitor module, and a solid-state thin-film lithium battery. The piezoelectric ceramic array is configured to harvest environmental vibration energy and store it in the supercapacitor module via a conversion circuit. The solid-state thin-film lithium battery is configured to start supplying power when the voltage of the supercapacitor module is lower than a first threshold. The dual-mode communication module includes a UWB positioning module and an RFID identification module.

[0011] The processing layer includes downhole edge nodes;

[0012] The core collaborative algorithm runs on the downhole edge node, including an energy consumption optimization submodule. This submodule is configured to execute the following hidden danger-energy consumption linkage control logic:

[0013] Real-time acquisition of vibration time-series data collected by the tag and real-time voltage of the supercapacitor module;

[0014] A lightweight one-dimensional dilated causal convolutional network is used to process vibration time series data, extract feature components in a preset frequency range, and determine whether there are precursors to the top plate pressing down.

[0015] Based on the coupling state between the judgment result and the real-time voltage, a working mode switching command is dynamically generated:

[0016] Under normal operating conditions, it switches between UWB full-function mode, RFID_ONLY mode and deep sleep mode according to the real-time voltage to achieve a self-sustaining balance between energy harvesting and consumption.

[0017] When the precursor to pressure on the top plate is detected, the UWB positioning module is forcibly locked into an active state and the positioning frequency is increased. At the same time, the power consumption of unnecessary modules other than UWB positioning is suppressed until the hidden danger is eliminated or the voltage drops to the second threshold.

[0018] Preferably, the piezoelectric ceramic array is made of lead zirconate titanate piezoelectric ceramics in parallel configuration, and the resonant frequency of the piezoelectric ceramic array is configured to match the vibration frequency of the underground coal mining machine, which is 40±5Hz.

[0019] The conversion circuit includes a Buck-Boost energy conversion circuit, configured to convert the collected mechanical energy into electrical energy and charge the supercapacitor module with an efficiency of not less than 75% in an environment with a vibration duty cycle ≥60%.

[0020] The supercapacitor module is configured to support the dual-mode communication module to operate at full capacity for ≥31 hours when fully charged, and to support the tag to maintain real-time clock operation when stationary.

[0021] Preferably, the explosion-proof rating of the energy-harvesting semi-passive dual-mode tag is Ex d [ib] I Mb;

[0022] The tag is configured with a triple energy limiting protection design: input voltage limiting uses a parallel bidirectional TVS diode to limit the breakdown voltage to the range of 3.3V-5V; energy storage current limiting uses a series current limiting resistor in conjunction with an electronic protection circuit to cut off the circuit within 1μs after a short circuit fault is detected; output energy limiting uses an intrinsically safe gate design to limit the internal inductance to <50μH and capacitance to <10μF, ensuring that the energy released under any fault condition is less than 20μJ.

[0023] Preferably, the lightweight one-dimensional dilated causal convolutional network includes 6 dilated convolutional layers with dilation coefficients set to [1,2,4,8,16,32] and a kernel size of 3, to construct a receptive field covering a time domain range of 1270ms, used to identify the 10-20Hz low-frequency vibration components in the precursor of the top plate pressure.

[0024] The energy consumption optimization submodule is configured to train the network using a weighted multi-objective loss function L, combined with the FocalLoss loss term and the MSE energy consumption prediction error term.

[0025] Where L = 0.6 × FocalLoss + 0.4 × MSE.

[0026] Preferably, the core collaborative algorithm further includes a collaborative verification submodule;

[0027] The collaborative verification submodule is configured to establish a multi-physical quantity coupled verification model of energy-trajectory-vibration, and use Kalman filtering to fuse the supercapacitor voltage drop rate and the positioning coordinate change rate of the tag.

[0028] When the tag is detected to be in a high-frequency emission active state but the slope of the supercapacitor voltage drop is significantly lower than the theoretical value and the deviation is >30%, or the positioning shows movement but the vibration monitoring shows that it is stationary and the deviation lasts for >3 seconds, it is determined that the data is abnormal or the tag has been illegally tampered with, and an abnormality warning is triggered.

[0029] Preferably, the transport layer is configured to use a post-quantum cryptography algorithm module for data encryption;

[0030] The post-quantum cryptography algorithm module is configured to use the CRYSTALS-Kyber key encapsulation mechanism and the Dilithium digital signature algorithm.

[0031] The core collaborative algorithm also includes an encryption-transmission collaborative submodule, which is configured to dynamically adjust the encryption strategy according to the data priority: for high-priority emergency and location data, a single-frame mode of CRYSTALS-Kyber-512 lightweight key encapsulation combined with ChaCha20-Poly1305 stream encryption is used; for ordinary-priority perception and tracing data, a batch mode of CRYSTALS-Kyber-768 standard key encapsulation combined with AES-256-GCM block encryption is used.

[0032] Preferably, the sensing layer further includes a reader / writer for recycling;

[0033] The recycling reader is configured to perform tag detection and disposal at the end of the entire life cycle: reading the reuse count, piezoelectric ceramic energy harvesting efficiency history, and chip self-test status stored inside the tag;

[0034] The system determines whether the tag meets the reuse conditions based on the read data. If the reuse conditions are met, the tag's supercapacitor is activated by radio frequency energy, and the EEPROM area ID is erased and a new ID is written. If the reuse conditions are not met, a Kill command or hardware reset signal is sent to put the tag into permanent sleep.

[0035] Preferably, the processing layer further includes a digital twin server;

[0036] The digital twin server is configured to build a high-fidelity 3D rendering model based on the Unity 3D engine, receive real-time positioning data, status data and energy consumption data distributed by the core collaborative algorithm, realize millisecond-level synchronous mapping between the underground physical scene and the virtual model, and overlay and display the real-time movement trajectory and health status indicators of the safety boot in the virtual model.

[0037] Preferably, the core collaborative algorithm is further configured to execute a full lifecycle tracing workflow, specifically including:

[0038] Data entry and binding steps: Read the tag ID with a reader and bind it with the physical information of the security boot, generate an initial traceability file containing the initial signature and store it on the data storage server;

[0039] Downhole real-time traceability steps: During the service period of the safety boot, the tag's positioning data, supercapacitor voltage fluctuation curve and vibration environment data are encrypted and signed using a post-quantum cryptography algorithm to generate an downhole trajectory chain with a timestamp and tamper-proof signature and a full-condition energy consumption record, forming a digital archive for the service period;

[0040] Emergency Enhancement Tracing Steps: When the energy consumption optimization submodule detects the precursor to the pressure on the top plate and forcibly locks the UWB positioning module, the system automatically increases the positioning data acquisition frequency and encrypts the storage to construct a high-density emergency spatiotemporal tracing segment for trajectory reconstruction and cause analysis after the accident.

[0041] Retirement and reuse traceability steps: After the safety boot is retired, the system reads the number of times the tag has been reused and the historical working condition data during its service period. Based on the modified Arrhenius-Coffin coupling model, the system calculates the cumulative lifespan loss and determines whether to reuse or scrap the boot based on the loss value and the reuse threshold. If reuse is determined, the system updates the ownership information and number of reuses in the traceability record and generates a new data signature. If scrapping is determined, the system seals the full lifecycle traceability data and destroys the tag ID.

[0042] Preferably, in the modified Arrhenius-Coffin coupling model, the equivalent life loss Ltotal = Lthermal + Lmechanical, where the thermal aging component Lthermal follows the Arrhenius equation, and the mechanical fatigue component Lmechanical follows the Coffin-Manson equation. The two are coupled through Miner's linear cumulative damage theory. The reuse threshold is set to 4-6 times, with a cumulative total service life of 36-54 months.

[0043] Preferably, the core collaborative algorithm further includes a model iteration submodule;

[0044] The model iteration submodule is configured to execute cloud-edge collaborative optimization logic: the downhole edge node is configured to run a lightweight model for real-time inference and upload the inference results and feature data to the ground cloud server; the ground cloud server is configured to use a large model to analyze and train the data, generate optimized lightweight model weights through knowledge distillation technology, and incrementally update them to the downhole edge node to adapt to the complex and ever-changing environmental characteristics of the downhole environment.

[0045] The beneficial effects of this invention are as follows:

[0046] First, the synergistic design of the composite energy storage architecture and triple energy limiting protection: A composite energy storage unit consisting of a PZT-5H piezoelectric ceramic array, a 10F supercapacitor, and a solid-state thin-film lithium battery was constructed, and a triple energy limiting protection circuit with TVS voltage limiting, resistor current limiting, and MOSFET fast cut-off (<1μs) was innovatively configured; this design achieves a net energy surplus of 2.38mW under continuous vibration environment, ensuring energy self-sufficiency for a single shift; more importantly, the energy released by the 36.45J large-capacity capacitor under fault conditions is strictly limited to 0.73μJ, which is far below the intrinsic safety threshold of 20μJ specified in GB 3836.4-2021.

[0047] Second, the algorithm-hardware collaboration of dilated causal convolution and dynamic state machine: A deep coupling algorithm-hardware collaborative architecture of dilated causal convolution (1D-TCN) and finite state machine (FSM) is proposed. By designing a 6-layer network with dilation coefficients of [1,2,4,8,16,32], a receptive field of 1270ms is constructed, specifically matching the top plate to suppress precursor cycles; at the same time, dynamic switching of high / medium / low energy three states (response <100ms) is realized based on the supercapacitor voltage Vcap; this architecture enables edge node inference latency <2ms, top plate suppression detection rate reaches 94.7%, and the average power consumption of the tag is controlled at 0.30mW, realizing cross-dimensional collaboration between hazard prediction and energy management.

[0048] Third, a dynamic service and reuse mechanism based on accelerated aging testing and the Arrhenius-Coffin coupling model: A dynamic reuse mechanism based on scientific evaluation was established. The equivalent life loss under temperature, humidity and vibration coupling conditions was calculated using the modified Arrhenius-Coffin model. Reuse was determined by combining multi-dimensional performance thresholds such as supercapacitor capacity retention (≥85%), PZT conversion rate (≥1.2V), and cosine similarity (≥0.85) based on MFCC feature extraction. This mechanism not only supports multiple reuses of tags, but also ensures the data credibility of reused tags through ID-reuse-sensing association verification (comparing the historical fingerprint database of the tag's triaxial vibration MFCC features and temperature response). Combined with rapid functional screening (false positive rate ≤0.2%), closed-loop management of the entire life cycle of electronic tags was realized.

[0049] Fourth, the catastrophic redundancy fusion of post-quantum cryptography and through-ground communication: It integrates the CRYSTALS-Kyber quantum-resistant cryptographic algorithm with 3kHz magnetic induction through-ground communication (TTE) technology. In extreme cases where the underground 5G-A network is interrupted due to disasters, the TTE module can penetrate 100m of rock layer with a link margin of 140.56dB under dry rock conditions to stably transmit vital signs and location data at a rate of 50bps (20 bytes / 3.2 seconds), thus constructing a dual protection system of quantum-resistant security and catastrophic communication, providing reliable technical support for emergency rescue of underground personnel. Detailed Implementation

[0050] The RFID-based full lifecycle traceability system for mining safety boots includes a sensing layer, a transmission layer, a processing layer, and a customized core collaborative algorithm with a three-layer serial architecture.

[0051] (a) Perception layer

[0052] The perception layer serves as the data acquisition and execution terminal, including: energy-harvesting semi-passive dual-mode tags, UWB base stations, explosion-proof dual-mode readers, recycling readers, positioning calibration equipment, tag testing equipment, and tag packaging machines; the specific technical solutions for each device are as follows:

[0053] (1-1) Energy Harvesting Semi-Passive Dual-Mode Tag

[0054] Built-in composite energy storage and acquisition unit, including:

[0055] Main energy storage: 10F / 2.7V supercapacitor module (composed of 5 2F / 2.7V cells connected in parallel, internal resistance ≤0.5Ω, total energy storage E max =0.5×10×2.7 2 =36.45J);

[0056] Backup energy storage: Solid-state thin-film lithium battery (10mAh / 3.7V, energy 133.2J, isolated from the supercapacitor by an ideal diode controller, only activated when vibration energy <0.5mW and supercapacitor voltage <1.8V, with a battery life of ≥4 hours).

[0057] Energy harvesting: Lead zirconate titanate piezoelectric ceramic (PZT-5H) array (3 pieces in parallel, total size 30mm×20mm×0.5mm, each piece with a 10g tungsten alloy mass, resonant frequency 40±5Hz, electromechanical coupling coefficient k) 31 ≈0.35), and configured with Buck-Boost energy conversion circuit (efficiency η≥75%);

[0058] Energy balance and range proof:

[0059] Actual measurements show that under the vibration environment of a coal mining machine (40Hz / 2.5g, vibration duty cycle ≥60%):

[0060] Average sampling power: P in =3×2mW×75%×60%=2.7mW (3 chips in parallel, conversion efficiency 75%, duty cycle correction);

[0061] System average power consumption: UWB module duty cycle 0.05% (100μs transmission every 200ms), average power consumption P UWB =500mW × 0.05% = 0.25mW; RFID intermittent listening power consumption P RFID =0.05mW; MCU sleep power consumption P MCU =0.02mW; Total power consumption P out =0.32mW; The 0.32mW is a purely theoretical calculation value (considering only the UWB duty cycle), and does not take into account the MCU dynamic power consumption and state switching loss;

[0062] Energy surplus: P surplus =2.7−0.32=2.38mW, with a cumulative surplus of 68.5J over 8 hours, far exceeding the supercapacitor capacity (36.45J). The system achieves energy self-sufficiency under continuous vibration.

[0063] Transient buffering and endurance: Under static conditions without energy replenishment, the supercapacitor's 36.45J energy can support the system to work continuously for approximately 36.45J / 0.32mW≈31.6 hours, which is much longer than the common vibration intervals in downhole (usually <30 minutes), ensuring that the UWB function is not interrupted during short shutdowns;

[0064] Dynamic service life and reuse mechanism:

[0065] Based on the modified Arrhenius-Coffin coupling model, the equivalent lifetime loss Ltotal = Lthermal + Lmechanical, where the thermal aging component Lthermal follows the Arrhenius equation and the mechanical fatigue component Lmechanical follows the Coffin-Manson equation. The two are coupled through Miner's linear cumulative damage theory.

[0066] Accelerated aging test: Under 85℃ / 85%RH / 2g vibration conditions for 1000 hours, the equivalent downhole 25℃ / 2g environment life loss is about 180 days, calculated by the Eyring model.

[0067] Adaptive Cycle: The single service cycle of the tag is not a fixed value, but is dynamically adjusted based on the edge node's assessment of downhole temperature, humidity, and vibration intensity.

[0068] Under normal conditions (temperature and humidity <60℃ / 80%): Recommended cycle is 9 months;

[0069] Harsh environments (>80℃ / 90%RH): Recommended cycle is 6 months;

[0070] Full life cycle closed loop: Set the safety reuse life threshold to 4-6 times, with a total cumulative service life of 36-54 months. By matching it with the life of safety boots (3-5 years), a full life cycle closed loop is achieved.

[0071] Intrinsically safe explosion-proof design:

[0072] Explosion-proof marking Ex d [ib] I Mb, employing triple energy limiting protection to ensure spark energy <20μJ:

[0073] Input voltage limiting: A bidirectional TVS diode (breakdown voltage 3.3V, clamping voltage 5V) is connected in parallel at the output of the PZT to prevent high-voltage arcing caused by open circuit;

[0074] Energy storage current limiting and rapid disconnection: Current limiting resistor R in series with the positive electrode of the supercapacitor limit =10Ω / 0.5W, used to limit fault current (I SC =0.27A) and instantaneous power (P max =0.73W); at the same time, in conjunction with the electronic protection circuit (MAX4173+DMN3018LSDMOSFET, response time <1μs), the circuit is cut off within 1μs after a short circuit fault is detected, limiting the actual spark energy released by the capacitor to the level of E =0.73×1=0.73uJ under fault conditions (far less than the 20μJ safety threshold).

[0075] Output energy limitation: The UWB / RFID radio frequency front end adopts an intrinsically safe barrier design (compliant with GB 3836.4-2021 Table A.1), with an internal inductance of <50μH and a capacitance of <10μF, ensuring that the energy released under any fault condition is <20μJ;

[0076] Energy management and power supply logic:

[0077] The system adopts a state machine management strategy based on supercapacitor voltage:

[0078] High energy state (V cap ≥2.5V): Enable UWB full-function mode, utilize stored charge to support the high instantaneous power consumption of the UWB module (transmission power consumption ≤500mW, transmission duration ≤100μs). At this time, the average power consumption of the system is controlled within 0.32mW, which is lower than the acquisition power of 2.7mW, thus achieving energy self-sufficiency.

[0079] Medium energy state (1.8V≤V) cap<2.5V): Automatically switches to RFID_ONLY mode, maintaining only area identification and reducing power consumption to wait for energy storage to recover;

[0080] Low energy state (V cap < 1.8V): Enters deep sleep mode, keeping only the RTC running;

[0081] This strategy ensures that the system can support continuous operation for ≥8 hours under continuous vibration energy harvesting conditions, meeting the requirements of AQ6210-2007 standard; under extreme static conditions, the backup lithium battery can maintain the system standby for ≥4 hours, ensuring emergency communication capabilities.

[0082] (1-2) UWB base station

[0083] Compliant with IEEE 802.15.4a standard, operating in the 6.5GHz band (channel 5), consistent with the frequency band of energy harvesting semi-passive dual-mode tag UWB chips; the UWB transmit peak power is 500mW, using pulse intermittent transmission (pulse width ≤2ns, repetition frequency 250kHz, duty cycle 0.05%), with an average transmit power of only 0.25mW; and through an intrinsically safe power limiter (series 2.2Ω current-limiting resistor + 3.3V Zener diode, located at the RF front end) to ensure average power <20mW and spark energy <20μJ under fault conditions, meeting the intrinsic safety requirements of GB 3836.4-2021 for RF equipment;

[0084] Hardware configuration: Built-in high-gain directional antenna (gain ≥12dBi, beamwidth 60°, front-to-back ratio ≥15dB) and bandpass filter module (center frequency 6.5GHz, bandwidth 500MHz, out-of-band rejection ≥40dBc); adopts mining explosion-proof design (explosion-proof rating Ex d I Mb, compliant with GB 3836.2-2010), metal shell material is Q235B steel, shell thickness ≥4mm, impact energy resistance ≥7J, suitable for underground 20-200Hz random vibration environment (acceleration ≥5g);

[0085] Core functions: Receives UWB pulse signals (single pulse width ≤ 2ns, repetition frequency 250kHz) transmitted by energy-harvesting semi-passive dual-mode tags; performs signal correlation calculations and TOA (Time of Arrival) estimation through edge preprocessing; uploads ranging results (not the original waveform) to the timing scheduling submodule every 50ms; supports multi-base station TDOA collaborative networking (≥4 base stations); base stations achieve time synchronization via the IEEE 1588v2 protocol (synchronization error ≤ 1ns, equivalent ranging error ≤ 30cm); networking latency ≤ 20ms; single base station coverage distance ≥ 50m (line-of-sight environment); after adopting NLOS error compensation algorithms (such as LSSVM or UKF), positioning accuracy reaches ≤ 10cm in line-of-sight environment and ≤ 30cm in non-line-of-sight environment; establishes a roadway geometric mapping model in conjunction with positioning calibration equipment (such as a laser total station); periodically corrects multipath errors to ensure no blind spots in the positioning signal;

[0086] (1-3) Explosion-proof dual-mode reader (dedicated to maintenance / emergency mode)

[0087] It features UWB+RFID dual-mode read / write capability, conforms to ISO 18000-6C (UHF RFID, 920-925MHz) and IEEE802.15.4a (UWB, 6.5GHz) standards; its explosion-proof rating is Ex d [ib] I Mb (explosion-proof and intrinsically safe composite), and its protection rating is IP68 (the antenna interface uses an explosion-proof gland seal); it is suitable for underground environments with high humidity (relative humidity 95%), high dust (concentration ≥1000mg / m³), and strong electromagnetic interference.

[0088] Core functions:

[0089] Maintenance and inventory mode: Read tag IDs and historical data via UHF RFID (reading distance ≤15m, underground roadway environment), reading latency ≤50ms / tag, supports multi-tag anti-collision reading (throughput ≥50 tags / second).

[0090] Emergency search mode: Activate UWB ranging function to perform two-way ranging (TWR) with the tag to assist in locating trapped personnel (accuracy ≤0.3m, distance ≤30m).

[0091] Tag Management and Edge Preprocessing: Receive control commands (such as tag sleep wake-up, reset, deactivation) issued by the processing layer and forward them to the tags through the RFID port; perform CRC verification, deduplication, and compression on the read data, temporarily store it in the local cache (capacity ≥ 10,000 records), and transmit it to the downhole data gateway through the intrinsically safe RS485 / fiber optic interface, and then upload it to the transmission layer through the gateway.

[0092] Routine real-time positioning is accomplished by a fixed UWB base station network, and this device only enables the UWB function in maintenance and emergency scenarios.

[0093] (1-4) Recycling Reader

[0094] It adopts a mining-grade explosion-proof and intrinsically safe design, with an explosion-proof marking of Ex d [ib] I Mb (explosion-proof enclosure Ex d, internal circuitry Ex ib), and a protection rating of IP65 (for ground recycling workshop environments).

[0095] Core functions support end-of-life management of tags: Reading the tag's unique ID and historical traceability data via UHF RFID (ISO 18000-6C) (identification distance 0.1-1m, accuracy ≥99.9%); reading the tag's internal storage of reuse counts, chip self-test status (e.g., memory bad block rate), and PZT-5H energy harvesting efficiency history; uploading the above data to the maintenance-reuse collaboration submodule via an intrinsically safe communication interface (RS485 / Ethernet), and receiving the judgment result; for tags determined to be reusable, activating the tag's supercapacitor via RFID radio frequency energy, performing EEPROM area ID erasure and new ID writing (erasure / write latency ≤100ms, including radio frequency energy setup time ≤30ms); for tags determined to be scrapped or have reached their reuse life (4-6 times), sending a Kill command (compliant with ISO 18000-6C). 18000-6C (requires 32-bit access password verification), or outputs a hardware reset signal (low level, lasting 100ms) to put the tag into permanent sleep state; fixed recycling line in ground maintenance workshop (IP65 protection), or portable operation at underground temporary recycling point (requires explosion-proof battery box).

[0096] (1-5) Positioning and calibration equipment

[0097] It complies with AQ 6210-2007 "General Technical Conditions for Coal Mine Underground Workers Management System" and GB 3836.2-2010 explosion-proof standard; it has a built-in high-precision positioning reference module (using a mine explosion-proof laser tracker or total station, with a positioning reference error of ≤2cm in underground environment and ≤0.5cm in surface laboratory environment) and a multi-path environment adaptation module;

[0098] Core functions: Establish an underground UWB positioning control network, calibrate UWB base station coordinates (correcting installation errors and displacements caused by tunnel deformation) and time synchronization errors; use NLOS error compensation algorithms (such as LSSVM or VBUKF) to correct ranging deviations caused by underground multipath effects;

[0099] Calibration process: The linkage timing scheduling submodule and collaborative verification submodule are used to perform batch calibration of UWB base stations in the entire mine during the monthly shutdown maintenance period or daily maintenance shift (the calibration time for a single base station is ≤5 minutes).

[0100] Technical indicators: After calibration, the system positioning accuracy is ≤10 cm in line-of-sight environment and ≤30 cm in non-line-of-sight environment, meeting the national standard requirements for underground personnel positioning; storing calibration parameters (base station coordinate correction amount, NLOS compensation coefficient) for optimizing the positioning algorithm in the model iteration sub-module; having the function of abnormal warning for calibration results (alarm when the base station displacement > 5 cm or the synchronization error > 1 ns);

[0101] (1 - 6) Tag detection device

[0102] Adapting to multi-level detection of energy-harvesting semi-passive dual-mode tags, with protection level IP67 (ground environment) / explosion protection level Ex ib I Mb (underground environment);

[0103] Quick function screening (time-consuming ≤5 s per tag): Detecting UWB / RFID chip communication response (handshake delay ≤50 ms is normal); Detecting the voltage at the supercapacitor terminal (≥2.0 V is normal, reflecting the instantaneous transmitable energy); Initial inspection of appearance integrity (coarse screening of crack or negative pressure tightness based on CNN vision recognition); Accuracy rate ≥99.8% (false judgment rate ≤0.2%);

[0104] Deep performance detection (time-consuming ≤120 s per tag, ground laboratory environment):

[0105] Chip sensitivity: Measuring the minimum wake-up power (threshold ≤ -18 dBm);

[0106] PZT-5H conversion rate: Excited by an electromagnetic vibration table (40 Hz / 2 g / 10 s), measuring the rectified output voltage (new tag reference ≥1.5 V, reuse determination ≥1.2 V, allowable attenuation ≤20%);

[0107] Supercapacitor capacity: Constant current discharge test (0.1C rate, 1A constant current), capacity retention rate ≥85% (new tag 10F, reuse determination ≥8.5F);

[0108] Solid-state battery health: Measuring the internal resistance by AC impedance spectroscopy (new tag <100 mΩ, reuse determination <200 mΩ);

[0109] Package airtightness: Pressurizing and maintaining pressure for 30 s by the pressure drop method, with leakage rate ≤5 Pa / s being qualified;

[0110] Intelligent determination and warning: Linking with the maintenance-reuse collaboration sub-module, comprehensively judging reuse / scrap based on the data of quick screening and deep detection; Triggering local audible and visual alarms (buzzer + LED) when the detection is unqualified, and uploading the data to the management system;

[0111] Environmental adaptability: Operating temperature -20℃ to +60℃, relative humidity ≤95% (non-condensing), suitable for the environment of coal mine surface maintenance workshops; however, if used for temporary underground detection points, only the rapid function screening mode should be activated, and deep detection should be completed on the surface after surfacing.

[0112] (1-7) Label Packaging Machine

[0113] It adopts mining-grade intrinsically safe (Ex ib I Mb) encapsulation technology, and its core function is to seal and encapsulate reusable labels;

[0114] Technical implementation: Mining-grade transparent epoxy resin composite material (dielectric constant 3.2-3.8, transmittance ≥85% at 6.5GHz) is selected, with an embedded silicone buffer layer (Shore hardness 30A) to ensure mechanical coupling between the PZT-5H ceramic sheet and the shell (vibration transmission efficiency ≥70%); the UWB / RFID antenna area adopts laser windowing + transparent silicone patch (thickness 2mm, insertion loss <0.5dB) to ensure wireless communication;

[0115] Packaging process parameters: Automated production line ≥ 10 pieces / minute (including PZT positioning, antenna assembly, vacuum encapsulation (vacuum degree ≥ 95kPa), curing, and testing); Curing process is 80℃ / 2h double curing;

[0116] Finished product label specifications after packaging: Protection level IP68 (GB / T 4208-2017, 72 hours in 1 meter water); Explosion-proof level Ex d [ib] I Mb (explosion-proof and intrinsically safe composite): The outer shell is made of Ex d I Mb explosion-proof type (compliant with GB 3836.2-2010, Q235B steel), and the internal circuit is made of Ex ib I Mb intrinsically safe type (compliant with GB 3836.4-2021, limiting spark energy release under fault conditions <20μJ). The connection between the PZT-5H piezoelectric sheet and the circuit is locally made of Ex m I Mb encapsulation type (compliant with GB3836.6-2021, epoxy thickness ≥3mm) for moisture protection; Double explosion protection is achieved through epoxy resin overall encapsulation and internal energy limitation, suitable for coal mine methane environment;

[0117] Mechanical and structural features: The PZT ceramic sheet is coupled to the epoxy shell through a silicone rubber buffer layer to form a bottom-flexible-side-rigid composite constraint structure; the UWB / RFID antenna area adopts laser windowing technology and is fitted with a wave-transparent silicone patch;

[0118] Performance indicators: The leakage rate of the encapsulated tag is ≤5Pa / s (the airtightness standard is consistent with the depth detection), the design life is 3 years, it can be reused 4 times, and the cumulative downhole use time is ≥24 months; after encapsulation, the RFID will be automatically triggered to write the initial ID and synchronized to the database storage server to complete the data entry and binding.

[0119] (ii) Transport Layer

[0120] The transport layer is responsible for secure data transmission, downlink command transmission, and emergency link protection. It includes: a post-quantum cryptography (PQC) algorithm module (the encryption scheme uses post-quantum cryptography (PQC) as the basic guarantee, and quantum key distribution (QKD) equipment can be optionally deployed in the ground core network as an enhancement to form a PQC+QKD hybrid encryption architecture), 5G-A macro base stations, 5G-A micro base stations, 5G-A explosion-proof repeaters (explosion-proof level Ex d I Mb, protection level IP68, single unit coverage distance ≥30m, relay transmission latency ≤5ms, supporting multipath routing algorithms), through-ground communication (TTE) emergency transceivers (deployed in underground refuge chambers), and mining satellite terminals (deployed at the ground wellhead).

[0121] The transport layer is uniformly scheduled by the core collaborative algorithm to achieve hierarchical transmission and multi-link parallel emergency backup. The specific working mode is as follows:

[0122] (2-1) Normal transmission mode

[0123] Data from the underground sensing layer is aggregated via a 5G-A wireless private network (macro base station + micro base station + repeater), and end-to-end encrypted using a post-quantum cryptography (PQC) algorithm (or AES encryption using QKD distributed keys). The data is then transmitted back to the ground core network via the underground industrial ring network / fiber optic cable, and finally uploaded to the cloud server. The single-link transmission latency is ≤20ms, which meets the requirements for real-time positioning and control.

[0124] (2-2) Emergency transmission mode (three-level redundancy mechanism)

[0125] The system automatically triggers the following emergency strategies based on the level of the fault:

[0126] Level 1 Response (Local Displacement / Obstruction in the Mine): When the signal of the original connected base station weakens (RSSI < -85dBm) or is interrupted, it automatically switches to the adjacent 5G-A micro base station or explosion-proof repeater, and maintains the communication link through a multipath routing algorithm to ensure uninterrupted data transmission.

[0127] Level 2 Response (5G-A Network Outage in Underground): When the underground wireless private network fails, activate the through-ground communication (TTE) emergency link; use magnetic induction communication at a frequency of 3kHz (wavelength 100km), based on Maxwell's equations, the skin depth of electromagnetic waves in a conductive medium is δ=[2 / (ωμσ)]. 1 / 2 For common shale or sandstone formations found downhole (electrical conductivity σ ranges from 0.01 to 0.1 S / m):

[0128] Skin depth calculation: With an operating frequency of 3kHz, the skin depth is approximately 92m in dry rock strata (σ=0.01S / m), with an attenuation of approximately 9.4dB at 100m; in moist rock strata (σ=0.1S / m), the skin depth is approximately 29m, with an attenuation of approximately 30dB at 100m; by increasing the transmit power (40dBm) and using high-sensitivity reception (-120dBm), 100m penetration communication can be achieved under dry rock strata conditions;

[0129] Link budget and penetration capability: Transmit power 10W (40dBm), loop antenna gain -10dBi, attenuation of 100m in rock (σ=0.01 S / m): 8.686×100 / 92≈9.44dB, receiver magnetic sensor sensitivity -120dBm, link margin = 40dBm (transmit) - 10dB (antenna loss) - 9.44dB (rock attenuation) - (-120dBm) (receiver sensitivity) = 140.56dB; even in wet rock (σ=0.1 S / m, attenuation about 30dB), there is still a link margin of more than 110dB, which meets the communication reliability requirements;

[0130] The underground transmitting unit is equipped with a loop antenna (2m in diameter, 100 turns, 4mm² copper wire), and the ground receiving station is equipped with a high-sensitivity magnetic sensor (noise level 0.1pT / √Hz). The transmission rate is 50bps (Manchester encoding), and it takes about 3.2 seconds to transmit 20 bytes of extremely simple data (vital signs code + location code). It directly penetrates the rock formation and transmits to the ground TTE receiving station, directly connecting to the rescue command center. The TTE module is equipped with an independent backup power supply (3.6V / 19Ah lithium-ion battery, standby current <50μA), supporting continuous transmission for ≥72 hours.

[0131] Level 3 Response (Ground Core Network External Connection Interruption): When underground data is successfully uploaded to the ground computer room, but the backbone fiber optic connection from the ground core network to the external cloud is interrupted, the algorithm will reroute the data stream to the wellhead satellite terminal, and then forward it to the cloud via the mining satellite communication system (such as Tiantong-1 or a high-throughput satellite with emergency backup link redundancy) to ensure that the entire mine's data is not lost under extreme disasters.

[0132] (2-3) Security Guarantee Mechanism

[0133] Quantum attack resistance: Employing NIST-standardized post-quantum cryptography (PQC) algorithms (such as CRYSTALS-Kyber key encapsulation mechanism and Dilithium digital signature algorithm), it effectively resists the threat of future quantum computers breaking traditional encryption algorithms;

[0134] Hybrid encryption architecture: QKD devices can be deployed to periodically distribute quantum keys to the 5G-A core network via optical fiber, realizing a hybrid encryption mode of quantum key + classical algorithm;

[0135] Low latency performance: The encryption module is accelerated and optimized based on the mining industrial-grade FPGA (XC7A35T) + ARM Cortex-M7 dual-core hardware architecture. The CRYSTALS-Kyber-512 key encapsulation time is ≤0.5ms, and the additional encryption latency after superimposed stream encryption is <1ms, which fully meets the stringent requirements of 5G-A uRLLC (ultra-reliable low-latency communication) scenarios.

[0136] (III) Processing Layer

[0137] The processing layer is responsible for data processing, analysis, storage, and instruction generation, including: downhole edge nodes, ground cloud servers, data storage servers, data synchronization gateways, and digital twin servers;

[0138] (3-1) Downhole edge node

[0139] Hardware configuration: computing power ≥1 TOPS (INT8 precision), built-in 128GB industrial-grade storage, suitable for explosion-proof environments in mines;

[0140] Core functions: Responsible for real-time shallow prediction of underground hazards (inference latency ≤80ms), local caching of raw data (72-hour cyclic overwrite, can survive network outages), and execution of real-time control commands issued from the cloud;

[0141] Software architecture: Runs lightweight neural networks (such as the MobileNetV3 architecture), supports the TensorFlow Lite inference framework, and achieves low-power, low-latency edge computing;

[0142] (3-2) Ground-based cloud server

[0143] Core capabilities: Deploy large-scale models specifically for mine safety (based on Transformer architecture or temporal neural networks), with elastic and scalable high-performance computing cluster resources;

[0144] In-depth analysis: Responsible for in-depth pattern analysis, including but not limited to hazard trend prediction, multi-source data correlation mining, and full lifecycle source tracing analysis;

[0145] Model optimization: It undertakes the training task of the model and uses knowledge distillation technology to transfer and compress the knowledge of the large model into a lightweight model, which is then distributed to the edge nodes to achieve efficient collaboration between cloud training and edge inference;

[0146] (3-3) Data storage server

[0147] Tiered architecture: Adopts a hot-cold data tiered storage strategy to optimize read / write performance and cost;

[0148] Hot data: Stores high-frequency access data from the past 7 days (such as real-time location and real-time alarms), using an NVMe SSD array to ensure high-concurrency read and write operations;

[0149] Cold data: Stores long-term data such as historical traceability records and tag reuse records, using distributed object storage (such as Ceph or HDFS), supporting petabyte-level data traceability and archiving;

[0150] (3-4) Data synchronization gateway

[0151] Deployed at the wellhead, it serves as the physical isolation and protocol conversion hub for data exchange between the downhole and the surface;

[0152] Core functions: Achieve physical security isolation between the underground industrial control network and the ground office / cloud network (using mining-specific one-way network gateway technology, compliant with mining network security specifications); support bidirectional conversion between underground industrial protocols (Modbus TCP, PROFINET, etc.) and ground cloud standard protocols (MQTT, HTTP, etc.); feature data compression (compression ratio ≥10:1), local caching, and breakpoint resume functionality to adapt to the unstable network environment in mines and ensure no data packet loss;

[0153] (3-5) Digital Twin Server

[0154] Engine architecture: Built on the Unity 3D engine architecture, supporting high-fidelity 3D rendering;

[0155] Real-time mapping: Receives real-time data uploaded by edge nodes (end-to-end latency ≤100ms) to achieve millisecond-level synchronous mapping between the underground physical scene and the virtual model;

[0156] Functional specifications: geometric accuracy 1:1, status synchronization delay ≤100ms, supporting immersive roaming, equipment status monitoring and emergency response simulation for management personnel;

[0157] (3-6) Cloud-edge collaboration mechanism

[0158] Data uplink: Edge nodes extract and upload feature data (not raw waveform data) to the cloud periodically (e.g., every hour) to reduce bandwidth usage;

[0159] Model delivery: The cloud dynamically adjusts the model delivery strategy based on the prediction accuracy of edge nodes.

[0160] When the accuracy is ≥99.8%, the optimized model weights are distributed every 2 hours.

[0161] When the accuracy is less than 99%, the frequency of updates is reduced to once every 30 minutes to quickly correct model biases.

[0162] Update technology: The distributed model weights are quantized and pruned, with a size of ≤5MB. They are quickly distributed to edge nodes for hot updates through the data synchronization gateway without restarting the device, thus realizing a closed loop of perception-prediction-optimization.

[0163] (iv) Core Cooperative Algorithm

[0164] The core collaborative algorithm is the core logic hub of the system, and its hardware carrier is the embedded chip of the downhole edge node and the ground cloud server;

[0165] (4-1) Embedded chip for downhole edge node

[0166] Select a mining-specific embedded AI chip (such as a high-performance, low-power SoC based on RISC-V or ARM architecture), with the specific compatibility parameters as follows:

[0167] Computing performance: Computing power ≥ 1 TOPS (INT8 precision), meeting the real-time inference requirements of lightweight 1D-TCN neural network and 7 sub-modules;

[0168] Electrical and Environmental Specifications: Operating voltage 3.3V, typical power consumption ≤5W, operating temperature (-40)℃ to (+85)℃ (industrial grade wide temperature range), shock resistant and adaptable to random vibration environments of 20-200Hz in underground wells (acceleration ≥5g);

[0169] Interface and Compatibility: Supports custom instruction set extensions, allowing direct execution of algorithm logic; and supports SPI / I... 2 The C interface reads the supercapacitor voltage of the PZT-5H piezoelectric ceramic energy harvesting module; and is compatible with data exchange with UWB / RFID chips (IEEE802.15.4a / ISO 18000-6C protocol stack);

[0170] Explosion-proof certification: The whole machine has passed the Ex ib I Mb (intrinsically safe) or Ex m I Mb (encapsulated) explosion-proof certification to ensure intrinsic safety in dangerous areas such as underground gas and coal dust;

[0171] (4-2) Ground-based cloud server

[0172] A high-performance server cluster specifically designed for mining operations was selected, with the following specific compatibility parameters:

[0173] Computing core: The CPU is a new generation of high-performance processors (such as Intel Xeon Gold or AMD EPYC series, supporting AVX-512 instruction set to accelerate AI inference and matrix operations).

[0174] Memory and storage: ≥128GB DDR4 ECC memory (supports high-concurrency data processing); configured with a 2TB NVMe SSD system disk (ensuring high-speed read and write of the system and model), and configured with a local high-speed cache disk (≥10TB SATA / SSD), while connecting to a distributed object storage cluster (Ceph / HDFS) to achieve persistent storage and tracing of PB-level historical data throughout the entire lifecycle;

[0175] Network capabilities: Dual 10 Gigabit Ethernet interfaces, supporting dual-mode access of 5G-A core network / satellite communication gateway; core network to edge node transmission latency ≤20ms, model parameter synchronization latency ≤30ms.

[0176] Business capacity: It can carry out the training and inference tasks of large-scale mining safety models (parameters ≥ 1B) in parallel, support real-time rendering and calculation of digital twin systems, and support hot update synchronization of model weights with underground edge node chips (seamless incremental update).

[0177] (4-3) Software logical architecture

[0178] Supported by the aforementioned hardware, the software logic consists of seven dedicated sub-modules: an energy consumption optimization sub-module (integrating hazard-energy consumption linkage judgment function), a timing scheduling sub-module, an encryption-transmission collaboration sub-module (integrating emergency priority judgment function), a data interaction sub-module, a model iteration sub-module, an instruction issuance sub-module, and a collaborative verification sub-module (integrating multimodal data cross-verification function). This architecture achieves deep integration of five major technologies: energy harvesting semi-passive IoT, UWB / RFID dual-mode positioning, post-quantum cryptography (PQC) and quantum key distribution (QKD) fusion encryption, 5G-A / satellite transmission, and digital twin. Through a cloud-edge collaboration mechanism of cloud training-edge inference, it realizes closed-loop management of the entire life cycle of the safety boot.

[0179] (4-3-1) Detailed functions of 7 dedicated sub-modules

[0180] (4-3-1-1) Energy consumption optimization submodule (integrated hidden danger-energy consumption linkage judgment function)

[0181] For compatibility with embedded chips in downhole edge nodes, a high-performance embedded AI chip specifically designed for mining (a high-energy-efficiency SoC based on RISC-V or ARM architecture) is selected. Specific compatibility parameters are as follows:

[0182] Computing performance: Built-in dedicated neural network processing unit (NPU) with computing power ≥1.2 TOPS (INT8 precision), meeting the real-time inference requirements of lightweight one-dimensional dilated causal convolutional network (1D-TCN) and multi-task concurrent processing, and reserving more than 20% computing power redundancy for future algorithm expansion; receives energy data (voltage, current) and supercapacitor energy storage data (SOC, terminal voltage) from PZT-5H piezoelectric ceramic sheet of energy harvesting semi-passive dual-mode tag every 10ms; when the continuous working time is >8 hours, triggers low power mode (RFID_ONLY) to ensure single-shift endurance; and when the early signs of pressure coming from the top plate are detected, locks the UWB full-function mode and increases the positioning frequency, while suppressing unnecessary power consumption to reserve transmission energy;

[0183] Model architecture and core algorithms:

[0184] A 1D-TCN is constructed using dilated causal convolution, and a specific receptive field is designed for the precursor features of the top plate pressure (low-frequency vibrations of 10-20Hz).

[0185] Network structure: There are 6 dilated convolutional layers in total (Conv1-Conv3 are depthwise separable convolutions: 7200×1→3600×16→1800×32→900×64, kernel size 9, stride 2; Conv4-Conv6 are standard dilated convolutions), with dilation coefficient d=[1,2,4,8,16,32];

[0186] Receptive field calculation: Based on the one-dimensional dilated convolution receptive field formula R=1+(k-1)∑d i ((k=3) to obtain the receptive field of 127 time points (corresponding to 1270ms time domain coverage), and through continuous processing by sliding window, identify the continuous abnormality of the 10-20Hz low frequency vibration component in the precursor of the top plate pressure.

[0187] Loss function: A weighted multi-objective loss function L is adopted to solve the problem of sparse hazard samples; where L = 0.6 × FocalLoss(hazard) + 0.4 × MSE(energy);

[0188] Network parameters: 23,425 total parameters, model size 23KB after INT8 quantization; ReLU activation function and Adam optimizer (learning rate 0.001) are used; measured inference latency on this NPU hardware platform is <2ms.

[0189] Model training dataset:

[0190] Data was collected from continuous monitoring at a mine from January to June 2023, with a time granularity of 10ms and a total sample size of 15.57 million. The ratio of normal operating condition samples to potential hazard samples was approximately 9:1 (positive samples included 800,000 samples of precursory roof collapse and 770,000 samples of equipment failure, while negative samples included 14 million samples of normal operating conditions). A weighted loss function was used to handle class imbalance. The normal operating condition samples totaled 14 million (vibration frequency 30-60Hz, amplitude <0.5g), the precursory roof collapse samples totaled 800,000 (vibration frequency 10-20Hz, amplitude >2g, labeled as roof collapse accidents occurring within 72 hours), and the equipment failure samples totaled 770,000 (classified as potential hazards). Labeling rules: roof collapse accidents or equipment failures occurring within 72 hours (acceleration >3g and duration >10s) were labeled as hazard label 1; otherwise, they were labeled as 0. Training configuration: Adam optimizer was used with an initial learning rate of 0.001 and a batch size of [missing information]. 128, trained for 50 epochs, validation set loss converged to 0.032; Model quantization: Post-training quantization (PTQ) was used with TensorFlow Lite, compressing the model size from 92KB to 23KB with an accuracy loss of <0.5%;

[0191] In actual measurement data from a certain mine from January to June 2023, compared with the conventional LSTM (control group), the 1D-TCN algorithm of this invention has a better energy consumption control strategy for tags when running on edge nodes:

[0192] In the measured data of a certain mine from January to June 2023, compared with the conventional LSTM (control group), the 1D-TCN algorithm of this invention, when running on the edge node, achieved a roof pressure detection rate of 94.7% (70.1% in the control group), a false alarm rate of 3.2 times / month (5.8 times / month in the control group), a tag-end power consumption of 0.30mW (0.52mW in the control group), and an inference latency of <2ms (12.7ms in the control group).

[0193] Wherein, 0.30mW is the average power consumption of the tag after the edge node dynamically adjusts the tag working mode according to this algorithm, which is lower than 0.52mW under the LSTM strategy; the 0.30mW is the weighted average power consumption of the tag under comprehensive working conditions (including 30% high energy state, 50% medium energy state, and 20% low energy state) under 1D-TCN optimized control.

[0194] (4-3-1-2) Timing Scheduling Submodule

[0195] Triggered by the energy optimization submodule, a scheduling instruction is generated within 5ms; it operates based on a finite state machine (FSM) architecture, with the core state monitored by the supercapacitor terminal voltage (V). capThe core states include UWB_ACTIVE (full-function activation), RFID_ONLY (RFID only), and DEEP_SLEEP (deep sleep).

[0196] Based on real-time operating conditions (hazard level, available energy characterization V) cap Dynamically adjust the chip's power domain and clock domain (device load rate):

[0197] High voltage state (V cap ≥2.5V, UWB_ACTIVE state: The PMU enables full-function power supply to the UWB RF front-end (6.5GHz) and RFID chip (920MHz), and the main control MCU runs at an 80MHz main frequency (using burst processing mode, with average power consumption controlled within 0.5mW); at this time, the energy acquisition rate (2.7mW) is still greater than the system's average consumption rate (estimated <1.0mW), and there is a continuous net energy surplus. The UWB acquisition frequency is dynamically adjusted according to personnel density (transmitting once every 200ms in high-density areas and once every 1000ms in low-density areas, with a constant pulse width of 100μs).

[0198] Medium voltage state (1.8≤V) cap <2.5V, RFID_ONLY state: The PMU cuts off the power supply to the UWB RF front end, retaining only the RFID chip for intermittent listening (wakes up once every 300ms, lasts for 10ms, average power consumption ≤0.05mW), the main control MCU enters low power mode or maintains low frequency operation (average power consumption corrected to 0.05mW), the total system power consumption drops to about 0.1mW, maintaining positive energy balance (2.7mW acquisition > 0.1mW consumption), waiting for Vcap to rise back to 2.6V;

[0199] Low voltage state (V cap <1.8V, DEEP_SLEEP state: Switches to deep sleep mode within 100ms, shuts down all RF modules, keeps only the RTC running (current <10μA, power consumption <0.03mW), prohibits communication activities with energy consumption exceeding the acquisition, and calculates the next minimum energy wake-up time by the energy consumption prediction algorithm (set hysteresis interval, when V cap Automatically wakes up when the voltage rises to 2.7V to prevent critical oscillation.

[0200] The multi-source data fusion module in the linkage processing layer completes data frame gap alignment at the moment of energy state switching (±5ms timestamp alignment), ensuring coordinated matching of energy, equipment, and acquisition timing, and avoiding data loss during state switching.

[0201] (4-3-1-3) Encryption-Transmission Collaboration Submodule (Integrated Emergency Priority Judgment Function)

[0202] Triggered by the data interaction submodule, the system intelligently adjusts data priority based on underground working conditions (gas concentration, personnel density, network congestion index) (Level 1: Emergency / Location; Level 2: Sensing / Tracing; Level 3: Log Statistics).

[0203] The control module for quantum cryptography (PQC) switches encryption modes based on data volume and priority.

[0204] Single-frame mode (high priority / small data volume <500 bytes): uses CRYSTALS-Kyber-512 lightweight key encapsulation + ChaCha20-Poly1305 stream encryption, end-to-end latency <1ms;

[0205] Batch mode (normal priority / large data volume): uses CRYSTALS-Kyber-768 standard key encapsulation + AES-256-GCM block encryption, GCM authentication tag 128bit;

[0206] If QKD devices are deployed, quantum-enhanced security can be achieved by replacing the PQC session key with the quantum key generated by QKD.

[0207] Integrated emergency priority judgment function: Based on real-time scene perception and pre-trained emergency mode model (trained offline based on 3-5 years of historical data, model size <10MB, inference latency <50ms), the priority weight is dynamically calculated within 100ms (e.g., the fault weight of high-risk area label is increased to 1.0); the queue depth and RSSI value of 5G-A base station are monitored in real time to predict link load and prioritize the allocation of idle links for Level 1 instructions; when the completion rate of Level 1 instruction processing is >80%, Level 2 instructions are preloaded into the transport layer cache to reduce handover latency;

[0208] The 5G-A macro / micro base stations in the scheduling transmission layer adapt to the network status (retransmission at reduced speed when RSSI <-85dBm, relay switching when RSSI <-95dBm) to achieve link load balancing and QoS guarantee.

[0209] (4-3-1-4) Data Interaction Submodule

[0210] By deploying a data synchronization gateway at the wellhead, a secure dedicated link is built between the edge node and the cloud and digital twin server based on a lightweight MQTT message bus (edge ​​side) and a cloud RabbitMQ cluster (based on IPSec / IKEv2 or TLS 1.3 encrypted tunnel, in compliance with the requirements of Information Security Level Protection 2.0).

[0211] The entire network clock is synchronized using the IEEE 1588v2 PTP protocol, ensuring that the clock synchronization error of distributed nodes is <1ms. Valid data received from the perception layer after collaborative verification is temporarily stored in a local circular buffer (capacity ≥10,000 records) at the edge node and transmitted uplink via MQTT over QUIC or 5G-A URLLC slicing. Cloud analysis results and control instructions are received and quickly distributed to the perception layer for execution. Distribution latency (edge ​​→ gateway one-way) is ≤10ms, and interaction latency (request-response round trip) is ≤15ms.

[0212] (4-3-1-5) Model Iteration Submodule

[0213] Triggered by the data interaction submodule, the self-adjusting iteration cycle is based on the shallow prediction accuracy of edge nodes (sliding window statistics, window length 1000 inferences) and data validity (completeness > 95%, outliers < 5%).

[0214] When the accuracy is ≥99.8% and the data validity meets the standard, the iteration cycle is 2 hours;

[0215] When the accuracy is <99% or the data anomaly rate is >5%, the iteration cycle is shortened to 30 minutes;

[0216] The parameters of the large model optimized in the cloud are lightweight and compressed (knowledge distillation + INT8 quantization + 50% structured pruning, compression time ≤1 minute, model size compressed from the original 20MB to ≤100KB), and then distributed to the edge nodes through the data synchronization gateway. After receiving the incremental update packet, the edge nodes use OTA technology for local hot reload (parameter memory overwrite), with a total transmission and loading latency of ≤500ms (based on the 2Mbps uplink bandwidth of 5G-A in the well), and it takes effect without restarting the device.

[0217] Record the iteration effects (accuracy changes, inference latency, energy consumption increments) to form a closed loop of cloud training - edge inference - effect feedback - cloud optimization;

[0218] The linked energy consumption optimization submodule dynamically adjusts the power budget threshold for switching operating modes based on the computational complexity (FLOPs) of the new model. Specifically, when the model complexity increases by 10% (corresponding to an increase of approximately 0.3mW in edge node power consumption), it automatically raises the threshold voltage for the UWB_ACTIVE state (e.g., from 2.5V to 2.6V) or extends the supercapacitor charging wait time to ensure that the energy self-sufficiency condition is not compromised.

[0219] (4-3-1-6) Command Issuance Submodule

[0220] Triggered by the data interaction submodule, cloud-based management / emergency commands (such as tag sleep, data acquisition start, forced deactivation) are converted into industrial control level signals (such as deactivation command: low level for 100ms, 3.3V TTL, drive current ≥10mA).

[0221] After being encrypted by the encryption-transmission coordination submodule, the data is transmitted to the perception layer via the 5G-A network or RS485 bus. Upon receiving execution feedback (ACK confirmation frame or status readback), the data is synchronized to the coordination verification submodule to verify the execution validity within the next 100ms cycle. After confirmation, the traceability data and digital twin model are updated (≤30ms).

[0222] (4-3-1-7) Collaborative verification submodule (integrates multimodal data cross-validation function)

[0223] (4-3-1-7) Collaborative verification submodule (integrates multimodal data cross-validation function)

[0224] Establish a multi-physical quantity coupling verification model of energy, trajectory, and vibration:

[0225] Kalman filtering is used for fusion, and the state vector X=[x, y, z, V] is... cap P harvest ] T Observation vector Z=[x UWB y RFID V measured ] T The core logic is to observe the voltage V of the supercapacitor. cap The actual power consumption of the tag is inferred from the rate of decline and compared with the reported activity status: if the positioning system reports that it is in a high-frequency transmission active state, but V cap If the rate of decline is significantly lower than the theoretical value (deviation > 30%), it is determined that the data is false or the label has been illegally tampered with.

[0226] After system startup, it runs continuously (startup latency ≤ 40ms, cross-validation performed every 100ms). It integrates multimodal data cross-validation functionality, establishes a localization-perception-source tracing correlation model through self-learning from historical data, employs lightweight temporal anomaly detection algorithms (such as Isolation Forest or LSTM prediction error method), and establishes a dynamic baseline by inputting multi-source data from the most recent 72 hours to achieve triple cross-validation.

[0227] Spatiotemporal consistency verification: This checks the matching between the rate of change of positioning coordinates and the amplitude of vibration sensors. Threshold settings: Movement is defined as vibration acceleration > 0.5g, and stationary state is defined as vibration acceleration < 0.5g. If positioning shows movement but vibration monitoring shows stationary, or vice versa, and the deviation lasts for > 3 seconds, an alarm is triggered.

[0228] Energy consumption-behavior consistency verification: verify the consistency between the supercapacitor voltage drop curve and the UWB / RFID transmission record to prevent false reporting (fake active signals) or energy theft (actual transmission but not reported).

[0229] ID-Reuse-Sensing Association Verification: Establish a historical sensing fingerprint database for tag IDs (extracting Mel-frequency cepstral coefficients (MFCCs) and temperature response characteristics from the tag's triaxial vibration spectrum). When a tag is reused, calculate the cosine similarity between the current sensing data and the fingerprint database. If the similarity is <0.85 (indicating a significant change in vibration characteristics), it is determined that the tag has been replaced or that an internal component has failed. If the number of reuses increases but the sensing data deviates from the fingerprint database (KL divergence >0.5 or exceeds the 3σ principle), it is also determined to be abnormal.

[0230] Intelligent anomaly type identification and closed-loop processing:

[0231] Minor deviations: Adjust the thresholds to match the system's high-precision specifications (line-of-sight < 0.2m, non-line-of-sight < 0.3m). Small errors within these ranges are automatically corrected using Kalman filtering or LSSVM to ensure data smoothness.

[0232] Severely invalid: If the positioning deviation exceeds the above range (line of sight > 0.2m or non-line of sight > 0.3m), the data integrity rate is < 90%, or the timestamp jump is > 200ms, it is judged as a severe anomaly and triggers re-acquisition (the current frame is discarded and new data is requested within ≤ 100ms).

[0233] Periodic self-calibration: Self-calibration is performed every 24 hours (during daily maintenance shifts) to comprehensively verify energy continuity, clock integrity, and positioning accuracy (refer to the standards in Chapters 1-2). Any abnormalities are fed back to the energy consumption optimization submodule or the timing scheduling submodule for parameter correction.

[0234] (4-3-2) Timing of 7 sub-modules

[0235] (4-3-2-1) Initialization phase: When the system is powered on or the miner goes down into the mine, the energy consumption optimization submodule (startup delay ≤ 50ms) is started first, collecting energy data (PZT-5H voltage / current) from the energy acquisition semi-passive dual-mode tag and supercapacitor energy storage data (SOC / terminal voltage) to establish an initial energy baseline; synchronously transmitting the data to the timing scheduling submodule (trigger ≤ 30ms), the timing scheduling submodule generates basic scheduling instructions based on the initial energy data; synchronously starting the collaborative verification submodule (startup ≤ 40ms), the entire process begins to verify all subsequent data and instructions;

[0236] (4-3-2-2) Data Acquisition Phase: When the energy consumption optimization submodule determines that the tag energy is ≥2mW, it triggers the timing scheduling submodule (response ≤20ms); the timing scheduling submodule starts the perception-positioning coordination logic, and the sensing layer device starts data acquisition; the hidden danger-energy consumption linkage judgment function in the energy consumption optimization submodule is started simultaneously, associating 72 hours of historical energy consumption and hidden danger data, generating energy consumption thresholds and acquisition frequency suggestions and feeding them back, and the energy consumption optimization submodule dynamically adjusts the positioning frequency and chip timing accordingly. After the data acquisition is completed, the data interaction submodule (latency ≤15ms) receives the preprocessed data, and the collaborative verification submodule verifies the acquired data every 100ms (including multimodal cross-verification). If the data is invalid, it triggers re-acquisition (completed within ≤100ms);

[0237] (4-3-2-3) Encrypted Transmission Phase: After the data interaction submodule receives valid collected data, it triggers the encryption-transmission coordination submodule (response ≤ 25ms); the emergency priority judgment function in the encryption-transmission coordination submodule is activated, and the data priority is determined within 100ms. The control quantum cryptography (PQC) processing module switches the encryption mode according to the priority, and at the same time, the transmission layer equipment is scheduled to adapt to the network status; the coordination verification submodule verifies the transmission integrity every 50ms; if the network signal is < -85dBm and lasts for ≥ 300ms, the encryption-transmission coordination submodule activates the emergency switching logic and switches to the underground emergency wired link (or explosion-proof Ethernet redundant ring network) within 300ms, and uploads the data to the explosion-proof satellite terminal at the ground wellhead, and transmits it via the satellite link as a backup;

[0238] (4-3-2-4) Data Processing Stage: After the encryption-transmission collaboration submodule completes data transmission, it triggers the data interaction submodule for secondary distribution (latency ≤ 15ms), distributing the decrypted data to the downhole edge nodes and the ground cloud server; the edge nodes complete shallow prediction within 80ms, and the cloud completes depth analysis within ≤ 1 minute, simultaneously triggering the model iteration submodule (response ≤ 20ms); the model iteration submodule adjusts the iteration cycle based on the prediction accuracy of the edge nodes and the analysis results of the cloud, and synchronizes the optimized parameters of the cloud to the edge nodes to update the model after lightweight processing; the collaborative verification submodule verifies the model iteration effect every 24 hours to ensure that the model accuracy meets the standards;

[0239] (4-3-2-5) Command Execution Phase: After data processing is completed, the data interaction submodule triggers the command issuance submodule based on the analysis results (response ≤20ms); the command issuance submodule converts the control / emergency command into an electrical signal that the device can recognize, and links the encryption-transmission coordination submodule to encrypt and issue the command to the perception layer device; after the perception layer device executes the command, it synchronizes the execution feedback to the coordination verification submodule (latency ≤10ms, including multimodal cross-verification of execution effect); the coordination verification submodule verifies the execution effect; after the verification is passed, the command issuance submodule updates the traceability data and digital twin model (≤30ms).

[0240] (4-3-2-6) Maintenance and Reuse Sequence: The instruction issuance submodule triggers the maintenance-reuse collaboration submodule at a preset cycle (2:00 AM every Monday / Tuesday / Wednesday) (response ≤25ms); the maintenance-reuse collaboration submodule schedules and locates calibration equipment and tag detection equipment, and regularly completes equipment calibration and tag performance testing (including ID-reuse-sensing association verification in multimodal verification), while linking with the energy consumption optimization submodule to update the energy threshold based on the test data; the collaboration verification submodule verifies the calibration, testing and reuse binding information to ensure compliance of the maintenance and reuse process; when tags are recycled, the maintenance-reuse collaboration submodule links with the recycling reader to complete tag deactivation, ID erasure and reuse eligibility determination;

[0241] (4-3-2-7) Model Optimization Phase: The model iteration submodule continuously schedules the data storage server and accumulates full-process data by category; the system calls the data to complete one model iteration according to an adaptive cycle (usually 1-2 hours, dynamically adjusted according to accuracy), optimizing the prediction threshold, data weight, and energy consumption parameters; after the optimized parameters are synchronized to the edge nodes, the energy consumption optimization submodule (including the hidden danger-energy consumption linkage judgment function) adjusts its working logic according to the new parameters to achieve self-adaptive optimization of the algorithm; the collaborative verification submodule (including the multimodal data cross-verification function) verifies the optimization effect throughout the process to ensure the stable operation of the algorithm;

[0242] (v) System lifecycle workflow

[0243] Using the core collaborative algorithm as the scheduling thread, closed-loop management of the entire lifecycle of the safety boot is achieved. The specific steps are as follows:

[0244] Step 1, database binding, details are as follows:

[0245] When an administrator triggers an inbound binding command, the algorithm initiates the tag binding process.

[0246] The explosion-proof dual-mode reader is dispatched to read the energy-harvesting semi-passive dual-mode tag ID and synchronously record the tag ID and safety boot information into the data storage server.

[0247] Preset tag erasure and rewrite parameters (≤4 times), and establish a mapping association between tag ID, security boot information, and reuse parameters;

[0248] The collaborative verification submodule verifies the integrity of the binding information. If it passes, it outputs a success signal; otherwise, it rereads the information. The time taken for a single binding is ≤10 seconds.

[0249] Step two, downhole deployment and data acquisition, are detailed below:

[0250] After the miner goes down into the mine, the tag energy acquisition module collects energy, and the algorithm starts the perception-localization collaborative logic;

[0251] The energy consumption optimization submodule is based on the supercapacitor voltage V. cap Threshold adjustment of positioning frequency and chip timing, while simultaneously coordinating with supercapacitor energy storage unit to dynamically distribute energy.

[0252] High energy state (V cap ≥2.5V): UWB uploads data every 100ms (average power consumption ≤0.5mW, lower than the acquisition power of 2.7mW to ensure energy self-sufficiency), the sensor collects data every minute, and the data is reduced by lightweight Kalman filter noise reduction. The instantaneous peak power of UWB transmission is provided by supercapacitor.

[0253] Medium energy state (1.8≤V) cap <2.5V): RFID data is collected every 300ms, and the sensor wakes up every 5 minutes;

[0254] Low energy state (V cap <1.8V): Switches to deep sleep mode within 100ms, shuts down all RF modules, keeps only the RTC running, and ensures a balanced power supply.

[0255] The multi-source data fusion module is scheduled to receive positioning data every 50ms, complete the filtering and weight allocation (UWB 0.8, RFID 0.2) within 20ms, and output positioning results ≤0.3m (line-of-sight ≤10cm, non-line-of-sight ≤30cm).

[0256] The scheduling chip transmits the processed data to the explosion-proof dual-mode reader / writer with a transmission latency of ≤100ms.

[0257] The collaborative verification submodule verifies the data every 100ms; if the data is invalid, it triggers a re-acquisition.

[0258] Step 3, secure data transmission, details are as follows:

[0259] The reader aggregates the data and transmits it to the post-quantum cryptography (PQC) processing module (supporting CRYSTALS-Kyber-512 / 768 algorithm), whereby the algorithm initiates the encryption-transmission coordinating submodule.

[0260] Hierarchical encrypted transmission:

[0261] High priority (location, emergency): single frame encryption (<1ms), 5G-A macro base station transmission, latency ≤20ms;

[0262] Medium priority (sensing, tracing): 10-frame batch encryption (≤3ms), 5G-A micro base station transmission, latency ≤50ms;

[0263] Low priority: 20-frame batch encryption, 5G-A repeater transmission, latency ≤100ms;

[0264] QKD Enhancement: If QKD devices are deployed (located in the ground core network), the quantum key generator generates a key every 10ms and distributes it to the underground edge node via optical fiber every 30 minutes to enhance the security of PQC session keys;

[0265] Emergency switching: The built-in emergency priority judgment unit monitors the network every 100ms. If the signal is lost for ≥300ms, it switches to the underground emergency link (explosion-proof Ethernet redundant ring network) within 300ms, uploads the data to the ground explosion-proof satellite terminal, and then transmits it via satellite; after the network is restored, it switches back to 5G-A mode and retransmits the data within 500ms.

[0266] The collaborative verification submodule checks the transmission integrity every 50ms; if the data is lost, it is retransmitted (≤50ms).

[0267] Step four, data processing and prediction / early warning, are as follows:

[0268] After the processing layer receives the decrypted data, the algorithm starts the data interaction submodule and distributes the data to the edge nodes and the cloud within 10ms;

[0269] Edge side: Receive data every 100ms, complete shallow prediction within 80ms (vibration ≥5mm / s and lasting ≥10s indicates excessive wear, positioning stillness ≥5min and no response indicates abnormal position), and synchronize to the cloud within 10ms;

[0270] Cloud-based: Perform in-depth analysis every minute and synchronize model optimization parameters to the algorithm every hour;

[0271] Model Iteration: The model iteration submodule synchronizes the lightweighted parameters to the edge nodes to update the model;

[0272] Digital twin: The system receives data every 500ms and performs a 1:1 real-time mapping (latency ≤100ms) for administrators to view;

[0273] The collaborative verification submodule verifies the model's performance every 24 hours (accuracy ≥ 99%), and if it fails to meet the standard, it iterates again.

[0274] Step 5, emergency response, details are as follows:

[0275] After the edge node predicts an accident, it transmits an emergency signal to the algorithm, and initiates the instruction distribution submodule and the emergency priority judgment unit.

[0276] Priority sorting: The emergency priority judgment unit completes the sorting within 100ms (Level 1: Personnel trapped, danger; Level 2: Tag and location failure; Level 3: Network and transmission anomaly).

[0277] Link assurance: The scheduling and transmission layer allocates links according to the corresponding priorities. Level 1 instructions are prioritized to be uploaded to the ground satellite terminal via the underground emergency wired link. The end-to-end transmission latency (edge ​​→ sensing layer) is ≤10ms, which is more than 50% shorter than the existing fixed priority scheme.

[0278] Emergency Response: A rescue plan is generated in the cloud within 1 second and synchronized to the digital twin system and rescue terminal within 100ms;

[0279] High-frequency data acquisition: The scheduling and perception layer collects emergency area data every 50ms and prioritizes its transmission to the digital twin system;

[0280] Command execution: The administrator issues a command, the algorithm converts it into a signal that the device can recognize and sends it out, and the execution result is fed back synchronously;

[0281] Closed loop: After the emergency response is completed, the algorithm stores the emergency data for model iteration;

[0282] Step six, maintenance, calibration, and disposal, are detailed below:

[0283] The algorithm is maintained every Monday, Tuesday, and Wednesday at 2 AM - reusing the collaborative process:

[0284] Regular calibration: Schedule equipment calibration, calibrate positioning parameters on Monday (≤10 minutes), check and optimize the transmission link on Tuesday, and calibrate the digital twin mapping accuracy on Wednesday (adjust when the error is ≥0.3m).

[0285] Scrap and Recycling: After the safety boots are scrapped, the tags are removed and placed into the recycling reader. The algorithm triggers the tag deactivation (low level for 100ms); the tag detection equipment is scheduled to complete two levels of detection.

[0286] (a) Rapid Function Screening (Time ≤ 5 seconds): RFID handshake test completed in the first 1-2 seconds (3 communications, success rate must reach 100%); EEPROM erase / write count N and current supercapacitor voltage read in the 3rd-4th seconds (≥ 2.0V is normal); Appearance AI inspection completed in the 5th second (identifying shell cracks based on CNN visual algorithm, accuracy ≥ 99%); False positive rate ≤ 0.2%;

[0287] (b) In-depth performance testing (time ≤ 120 seconds, ground laboratory environment): Chip sensitivity test (gradually reduce the RFID reader's transmission power, record the tag's minimum response power P min, pass / fail standard:

[0288] min≤−15dBm (new label baseline value -18dBm, allowable attenuation 3dB, corresponding performance retention ≥80%)); PZT-5H conversion rate test (electromagnetic vibration table excitation 40Hz / 2g, measuring rectified output voltage, pass standard: ≥1.2V); supercapacitor capacity test (0.1C rate constant current discharge, pass standard: capacity retention ≥85%, corresponding reuse threshold ≥8.5F); packaging hermeticity test (voltage drop method, -60kPa holding pressure for 30 seconds, pass standard: leakage rate ≤5Pa / s);

[0289] Reusability criteria: Chip performance ≥80%, energy harvesting efficiency ≥70%, and number of erase / write cycles N < 4 times; reusable tags can be re-bound after being encapsulated and erased, while non-reusable tags will enter permanent sleep mode by sending a Kill command (requires 32-bit password verification) or a hardware reset signal (low level for 100ms);

[0290] The collaborative verification submodule verifies calibration, testing, and reuse information to complete the closed loop.

[0291] Step 7, model optimization, as follows:

[0292] The algorithm initiates the model iteration submodule and schedules the data storage server to classify and accumulate data throughout the entire process.

[0293] Iteration cycle: The system calls the data iterative optimization model according to the adaptive cycle (usually 1-2 hours) to adjust the prediction threshold, data weight, and energy consumption parameters;

[0294] Parameter synchronization: After optimizing and lightweighting the parameters, they are synchronized to the edge nodes to update the model;

[0295] Self-adaptation: The algorithm adjusts the sampling frequency and transmission priority according to the optimized parameters to achieve self-adaptation of energy consumption and performance;

[0296] Performance verification: The model performance is verified every 24 hours, and if it does not meet the standards, it is iterated again.

Claims

1. A full lifecycle traceability system for mining safety boots based on RFID, characterized in that, It includes the perception layer, transmission layer, processing layer, and core collaborative algorithm; The sensing layer includes an energy-harvesting semi-passive dual-mode tag deployed on a mining safety boot. The tag includes a composite energy storage and harvesting unit and a dual-mode communication module. The composite energy storage and harvesting unit includes a piezoelectric ceramic array, a supercapacitor module, and a solid-state thin-film lithium battery. The piezoelectric ceramic array is configured to harvest environmental vibration energy and store it in the supercapacitor module via a conversion circuit. The solid-state thin-film lithium battery is configured to start supplying power when the voltage of the supercapacitor module is lower than a first threshold. The dual-mode communication module includes a UWB positioning module and an RFID identification module. The processing layer includes downhole edge nodes; The core collaborative algorithm runs on the downhole edge node, including an energy consumption optimization submodule. This submodule is configured to execute the following hidden danger-energy consumption linkage control logic: Real-time acquisition of vibration time-series data collected by the tag and real-time voltage of the supercapacitor module; A lightweight one-dimensional dilated causal convolutional network is used to process vibration time series data, extract feature components in a preset frequency range, and determine whether there are precursors to the top plate pressing down. Based on the coupling state between the judgment result and the real-time voltage, a working mode switching command is dynamically generated: Under normal operating conditions, it switches between UWB full-function mode, RFID_ONLY mode and deep sleep mode according to the real-time voltage to achieve a self-sustaining balance between energy harvesting and consumption. When the precursor to pressure on the top plate is detected, the UWB positioning module is forcibly locked into an active state and the positioning frequency is increased. At the same time, the power consumption of unnecessary modules other than UWB positioning is suppressed until the hidden danger is eliminated or the voltage drops to the second threshold.

2. The RFID-based full lifecycle traceability system for mining safety boots according to claim 1, characterized in that, The piezoelectric ceramic array uses lead zirconate titanate piezoelectric ceramics in parallel configuration, and the resonant frequency of the piezoelectric ceramic array is configured to match the vibration frequency of the underground coal mining machine, which is 40±5Hz. The conversion circuit includes a Buck-Boost energy conversion circuit, configured to convert the collected mechanical energy into electrical energy and charge the supercapacitor module with an efficiency of not less than 75% in an environment with a vibration duty cycle ≥60%. The supercapacitor module is configured to support the dual-mode communication module to operate at full capacity for ≥31 hours when fully charged, and to support the tag to maintain real-time clock operation when stationary.

3. The RFID-based full lifecycle traceability system for mining safety boots according to claim 1, characterized in that, The explosion-proof rating of the energy-harvesting semi-passive dual-mode tag is Ex d [ib] I Mb; The tag is configured with a triple energy limiting protection design: input voltage limiting uses a parallel bidirectional TVS diode to limit the breakdown voltage to the range of 3.3V-5V; energy storage current limiting uses a series current limiting resistor in conjunction with an electronic protection circuit to cut off the circuit within 1μs after a short circuit fault is detected; output energy limiting uses an intrinsically safe gate design to limit the internal inductance to <50μH and capacitance to <10μF, ensuring that the energy released under any fault condition is less than 20μJ.

4. The RFID-based full lifecycle traceability system for mining safety boots according to claim 1, characterized in that, The lightweight one-dimensional dilated causal convolutional network includes 6 dilated convolutional layers with dilation coefficients set to [1,2,4,8,16,32] and a kernel size of 3, to construct a receptive field covering a time domain range of 1270ms, used to identify the 10-20Hz low-frequency vibration components in the precursor of the top plate pressure. The energy consumption optimization submodule is configured to train the network using a weighted multi-objective loss function L, combined with the FocalLoss loss term and the MSE energy consumption prediction error term. Where L = 0.6 × FocalLoss + 0.4 × MSE.

5. The RFID-based full lifecycle traceability system for mining safety boots according to claim 1, characterized in that, The core collaborative algorithm also includes a collaborative verification submodule; The collaborative verification submodule is configured to establish a multi-physical quantity coupled verification model of energy-trajectory-vibration, and use Kalman filtering to fuse the supercapacitor voltage drop rate and the positioning coordinate change rate of the tag. When the tag is detected to be in a high-frequency emission active state but the slope of the supercapacitor voltage drop is significantly lower than the theoretical value and the deviation is >30%, or the positioning shows movement but the vibration monitoring shows that it is stationary and the deviation lasts for >3 seconds, it is determined that the data is abnormal or the tag has been illegally tampered with, and an abnormality warning is triggered.

6. The RFID-based full lifecycle traceability system for mining safety boots according to claim 1, characterized in that, The transport layer is configured to use a post-quantum cryptography algorithm module for data encryption; The post-quantum cryptography algorithm module is configured to use the CRYSTALS-Kyber key encapsulation mechanism and the Dilithium digital signature algorithm. The core collaborative algorithm also includes an encryption-transmission collaborative submodule, which is configured to dynamically adjust the encryption strategy according to the data priority: high-priority emergency and location data adopts a single-frame mode of CRYSTALS-Kyber-512 lightweight key encapsulation combined with ChaCha20-Poly1305 stream encryption, while ordinary priority perception and tracing data adopts a batch mode of CRYSTALS-Kyber-768 standard key encapsulation combined with AES-256-GCM block encryption.

7. The RFID-based full lifecycle traceability system for mining safety boots according to claim 1, characterized in that, The sensing layer also includes a reader / writer for recycling; The recycling reader is configured to perform tag detection and disposal at the end of the entire life cycle: reading the reuse count, piezoelectric ceramic energy harvesting efficiency history, and chip self-test status stored inside the tag; The system determines whether the tag meets the reuse conditions based on the read data. If the reuse conditions are met, the tag's supercapacitor is activated by radio frequency energy, and the EEPROM area ID is erased and a new ID is written. If the reuse conditions are not met, a Kill command or hardware reset signal is sent to put the tag into permanent sleep.

8. The RFID-based full lifecycle traceability system for mining safety boots according to claim 1, characterized in that, The processing layer also includes a digital twin server; The digital twin server is configured to build a high-fidelity 3D rendering model based on the Unity 3D engine, receive real-time positioning data, status data and energy consumption data distributed by the core collaborative algorithm, realize millisecond-level synchronous mapping between the underground physical scene and the virtual model, and overlay and display the real-time movement trajectory and health status indicators of the safety boot in the virtual model.

9. A full lifecycle traceability system for mining safety boots based on RFID according to any one of claims 1-8, characterized in that, The core collaborative algorithm is also configured to execute a full lifecycle tracing workflow, specifically including: Data entry and binding steps: Read the tag ID with a reader and bind it with the physical information of the security boot, generate an initial traceability file containing the initial signature and store it on the data storage server; Downhole real-time traceability steps: During the service period of the safety boot, the tag's positioning data, supercapacitor voltage fluctuation curve and vibration environment data are encrypted and signed using a post-quantum cryptography algorithm to generate an downhole trajectory chain with a timestamp and tamper-proof signature and a full-condition energy consumption record, forming a digital archive for the service period; Emergency Enhancement Tracing Steps: When the energy consumption optimization submodule detects the precursor to the pressure on the top plate and forcibly locks the UWB positioning module, the system automatically increases the positioning data acquisition frequency and encrypts the storage to construct a high-density emergency spatiotemporal tracing segment for trajectory reconstruction and cause analysis after the accident. Retirement and reuse traceability steps: After the safety boot is retired, the system reads the number of times the tag has been reused and the historical working condition data during its service period. Based on the modified Arrhenius-Coffin coupling model, the system calculates the cumulative lifespan loss and determines whether to reuse or scrap the boot based on the loss value and the reuse threshold. If reuse is determined, the system updates the ownership information and number of reuses in the traceability record and generates a new data signature. If scrap is determined, the system seals the full lifecycle traceability data and destroys the tag ID.

10. A full lifecycle traceability system for mining safety boots based on RFID according to claim 9, characterized in that, In the modified Arrhenius-Coffin coupling model, the equivalent life loss Ltotal = Lthermal + Lmechanical; where the thermal aging component Lthermal follows the Arrhenius equation, and the mechanical fatigue component Lmechanical follows the Coffin-Manson equation, and the two are coupled through Miner's linear cumulative damage theory; the reuse threshold is set to 4-6 times, with a cumulative total service life of 36-54 months.