Safety monitoring method for subway tunnel operation and intelligent safety helmet

Abstract

The application discloses a kind of subway tunnel operation safety monitoring method and intelligent safety helmet, it is related to subway operation and maintenance technical field, wherein, the safety monitoring method of subway tunnel operation includes the following monitoring steps: S100, obtains the intelligent safety helmet of multi-source perception fusion;S200, through the multi-source sensor array integrated on intelligent safety helmet, collect multi-source data information in tunnel environment, wherein, the multi-source data information includes laser ranging data, visual image data and environmental gas data;S300, utilize six-axis inertial measurement unit on intelligent safety helmet to collect the high-frequency motion trajectory data in the process of wearer walking, and through the high-frequency motion trajectory data, the laser ranging data and visual image data are carried out reverse geometric compensation;The application can carry out reverse geometric compensation to laser and image, realize in "dynamic walking" working condition, still can output high definition, no ghost tunnel wall surface data information.

Description

Technical Field

[0001] This invention relates to the field of subway operation and maintenance technology, and in particular to a safety monitoring method for subway tunnel operations and a smart safety helmet. Background Technology

[0002] In the entire lifecycle operation and maintenance of subway tunnels, traditional detection methods (such as manual inspection vehicles and large scanning equipment) suffer from difficulties in deployment, large blind spots, and poor real-time performance. Although basic smart safety helmets exist for subway tunnel operation and maintenance, most of these helmets only have positioning functions and generally suffer from the following three main problems in complex tunnel environments: First, unstable data acquisition; the sensor's posture swaying caused by the wearer's walking movements leads to severe motion artifacts in the data. Second, synchronization failures exist; data from sensors of different frequencies (LiDAR, IMU, cameras) lack a unified timestamp, making it impossible to accurately reconstruct the physical scene. Third, the interaction is limited; in high-noise and low-light environments inside tunnels, conventional buzzer alarms are easily drowned out by ambient noise, lacking a combined visual and tactile warning system.

[0003] This application proposes a safety monitoring method and a smart safety helmet for subway tunnel operations, which are used to solve the problems of inaccurate data collection during tunnel operation and maintenance, and the difficulty of effective interaction between the smart safety helmet and the wearer in the complex environment of the tunnel. Summary of the Invention

[0004] The main objective of this invention is to propose a safety monitoring method and a smart safety helmet for subway tunnel operations, aiming to solve the problems of inaccurate data collection and difficulty in effective interaction between the safety helmet and the wearer in the complex environment of the tunnel during tunnel operation and maintenance.

[0005] To achieve the above objectives, the safety monitoring method for subway tunnel operations proposed in this invention includes the following monitoring steps: S100, a smart safety helmet that acquires multi-source sensing fusion; S200: The multi-source sensor array integrated on the smart safety helmet collects multi-source data information in the tunnel environment, including laser ranging data, visual image data, and ambient gas data. S300: The six-axis inertial measurement unit on the smart safety helmet collects high-frequency motion trajectory data of the wearer during walking, and performs inverse geometric compensation on the laser ranging data and visual image data using the high-frequency motion trajectory data; S400, based on the hardware interrupt triggering mechanism in the smart safety helmet, performs millisecond-level time alignment on sensor data with different sampling frequencies and performs spatial mapping in a unified coordinate system to construct a fused space-time data cube; S500. Based on the space-time data cube, establish an environmental threshold comparison model, and perform weighted calculations by comprehensively considering the abnormal information of the lining surface structure, gas concentration information, and vibration intensity information to generate a risk index. S600. When the risk index exceeds the preset safety threshold, a warning is issued to the operator through a combination of visual, tactile and auditory modalities.

[0006] In one embodiment, S300 specifically includes the following steps: S310, physical first-order shock absorption is achieved through the flexible damping shock absorption structure between the inner lining of the smart safety helmet and the sensor module. S320: The six-axis inertial measurement unit acquires the operator's attitude angular velocity and acceleration data in real time at a sampling rate higher than that of the acquisition sensor. S330. Based on the attitude angular velocity and acceleration data, perform inverse geometric compensation on the interval jitter frequency caused by the person's walking frequency in the laser ranging data and visual image data, and output stable data to remove motion artifacts.

[0007] In one embodiment, S400 specifically includes the following steps: S410, at the bottom layer of the field programmable gate array accelerator, uses the interrupt triggering method of the synchronous bus to force a unified hardware timestamp on each frame of data acquired by the laser rangefinder, vision sensor, inertial measurement unit and gas sensor. S420. Based on the hardware timestamp, perform millisecond-level time alignment on sensor data with different sampling frequencies; S430: Map the aligned sensor data to a unified spatial coordinate system to generate a space-time data cube containing location, time, and multi-dimensional sensing information.

[0008] In one embodiment, in step S500, the formula for calculating the risk index is: R=W1×Δ D + W2×G +W3×V; Where R is the risk coefficient, Δ D , where is the abrupt change in laser ranging, characterizing the degree of structural anomaly of cracks or voids on the lining surface; G is the ambient gas concentration value; V is the ambient vibration intensity value; W1, W2, and W3 are the weighting coefficients corresponding to each index.

[0009] In one embodiment, in S500, the determination of the abnormal information of the lining surface structure includes the following determination conditions: When the laser reflectivity of the lining surface is abnormal or the ranging value changes abruptly, it is determined that there is a risk of cracks or voids. When the gas concentration exceeds the preset safety boundary value, it is determined that there is a risk of excessive harmful gas.

[0010] In one embodiment, S600 specifically includes the following steps: S610. When the risk index exceeds the preset safety threshold, a visual warning is given by flashing red light strip on the brim of the smart safety helmet; and, a tactile vibration is applied to the worker's head by a vibration feedback device that fits into the inner lining of the smart safety helmet; and, a voice warning is transmitted to the worker by a bone conduction speaker.

[0011] In one embodiment, the smart safety helmet integrates an analysis and control unit, and the anomaly detection, risk assessment, and early warning decision-making in S500 and S600 are all completed under the analysis and control of the analysis and control unit of the smart safety helmet.

[0012] This invention also proposes a smart safety helmet for implementing the aforementioned safety monitoring method for subway tunnel operations, the smart safety helmet comprising: The helmet itself; The sensing unit is located at the front end of the brim of the safety helmet body, and includes a six-axis inertial measurement unit, a laser ranging array, a wide dynamic range vision sensor, and an acoustic environment monitoring unit. The processing unit, embedded inside the helmet body, includes a low-power edge computing board. This board integrates a programmable gate array accelerator for performing sensor data fusion, pose compensation, and risk assessment. The interactive unit includes a light strip disposed around the periphery of the safety helmet body, a vibration feedback device fitted into the inner lining of the safety helmet body, and a bone conduction speaker, used to output multimodal warning information to the worker.

[0013] In one embodiment, a shock-absorbing module is provided on the inner lining of the safety helmet body. The shock-absorbing module includes a flexible damping bracket made of silicone shock-absorbing material, which is used to absorb the vibration generated by the movement of the worker's head and realize physical vibration isolation of the sensor module.

[0014] In one embodiment, the processing unit and the sensing unit are respectively located on the front and rear sides of the helmet body.

[0015] This invention enables inverse geometric compensation of laser and image data during environmental information acquisition. This allows for the output of high-definition, ghost-free tunnel wall data even under "dynamic walking" conditions, solving the problem of inaccurate data acquisition. Furthermore, the use of "spatiotemporal decoupling" technology allows for the fusion of data into an extremely clean data cube, ensuring millimeter-level accuracy in locating defects such as cracks and leaks. Moreover, this invention relies on local decoupling and risk analysis, controlling the entire decision-making process latency to the millisecond level. It can operate autonomously even without network connectivity. Simultaneously, by combining visual, tactile, and auditory sensors, it ensures that early warning information is delivered to workers in a completely closed-loop manner, maximizing the safety of frontline workers and demonstrating promising application prospects. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0017] Figure 1 A control flowchart of an embodiment of the safety monitoring method for subway tunnel operations provided by the present invention; Figure 2 To achieve Figure 1 A schematic diagram of the structure of an embodiment of a smart safety helmet for safety monitoring methods in subway tunnel operations.

[0018] Explanation of icon numbers: 1. Helmet body; 11. Sensing unit; 111. Six-axis inertial measurement unit; 112. Laser ranging array; 113. Wide dynamic range vision sensor; 114. Acoustic environment monitoring unit; 12. Processing unit; 121. Low-power edge computing board; 13. Interaction unit; 131. Light strip; 132. Vibration feedback device; 133. Bone conduction speaker; 14. Shock absorption module; 15. Adjustable strap; 16. Battery structure.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0022] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and / or" or "and / or" throughout the text includes three parallel solutions. For example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied simultaneously. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0023] In the entire lifecycle operation and maintenance of subway tunnels, traditional detection methods (such as manual inspection vehicles and large scanning equipment) suffer from difficulties in deployment, large blind spots, and poor real-time performance. Although basic smart safety helmets exist for subway tunnel operation and maintenance, most of these helmets only have positioning functions and generally suffer from the following three main problems in complex tunnel environments: First, unstable data acquisition; the sensor's posture swaying caused by the wearer's walking movements leads to severe motion artifacts in the data. Second, synchronization failures exist; data from sensors of different frequencies (LiDAR, IMU, cameras) lack a unified timestamp, making it impossible to accurately reconstruct the physical scene. Third, the interaction is limited; in high-noise and low-light environments inside tunnels, conventional buzzer alarms are easily drowned out by ambient noise, lacking a combined visual and tactile warning system.

[0024] This application proposes a safety monitoring method and a smart safety helmet for subway tunnel operations, which are used to solve the problems of inaccurate data collection and difficulty in effective interaction between the safety helmet and the wearer in the complex environment of the tunnel during tunnel operation and maintenance.

[0025] This invention proposes a safety monitoring method for subway tunnel operations.

[0026] Please see Figure 1 In one embodiment of the present invention, the safety monitoring method for subway tunnel operations includes the following monitoring steps: S100, acquiring a smart safety helmet with multi-source sensing fusion; S200, collecting multi-source data information in the tunnel environment through a multi-source sensor array integrated on the smart safety helmet, wherein the multi-source data information includes laser ranging data, visual image data, and ambient gas data; S300, collecting high-frequency motion trajectory data of the wearer during walking using a six-axis inertial measurement unit 111 on the smart safety helmet, and performing analysis on the laser ranging data and visual image data using the high-frequency motion trajectory data. Reverse geometric compensation; S400, Based on the hardware interrupt triggering mechanism in the smart safety helmet, perform millisecond-level time alignment on sensor data with different sampling frequencies, and perform spatial mapping in a unified coordinate system to construct a fused space-time data cube; S500, Based on the space-time data cube, establish an environmental threshold comparison model, and perform weighted calculation by comprehensively considering abnormal information of lining surface structure, gas concentration information and vibration intensity information to generate a risk index; S600, When the risk index exceeds the preset safety threshold, issue a warning to the workers through a combination of visual, tactile and auditory modalities.

[0027] The "multi-source sensing fusion smart safety helmet" refers to a smart safety helmet that integrates various types of sensors, edge computing units, and multimodal early warning devices on the basis of a conventional industrial smart safety helmet. Multi-source sensing fusion emphasizes that the smart safety helmet simultaneously carries sensors based on different physical principles (including optical, inertial, and gas detection sensors), and fuses the heterogeneous information output by each sensor into a unified data representation form through data processing algorithms. The "multi-source sensor array" refers to a collection of multiple sensor modules integrated in an array layout at the front of the smart safety helmet's brim, including but not limited to a laser rangefinder, a wide dynamic range vision sensor 113, a six-axis inertial measurement unit 111, and an acoustic environment monitoring unit 114. The "laser rangefinder data" refers to the distance information obtained by the laser rangefinder by emitting laser pulses towards the tunnel wall and receiving the reflected signals, used to characterize the spatial geometric contour of the tunnel lining surface. The "visual image data" refers to the wall image information collected by the wide dynamic range vision sensor 113 in the dark environment of the tunnel, used to identify visual defects such as water seepage, cracks, and spalling on the lining surface. The "ambient gas data" refers to the real-time detection and output of the concentration of harmful gases (such as methane, hydrogen sulfide, and carbon monoxide) in the tunnel space by the gas sensor. The six-axis inertial measurement unit 111 (IMU) is an inertial sensor device integrating a three-axis gyroscope and a three-axis accelerometer, capable of outputting the wearer's three-dimensional angular velocity and three-dimensional linear acceleration of the head in real time. The "high-frequency motion trajectory data" refers to the continuous time-series data of the wearer's motion state acquired by the IMU at a high sampling frequency (typically higher than that of laser and vision sensors). The "reverse geometric compensation" refers to using the motion data captured by the IMU to calculate the sensor's pose offset at the moment of acquisition and to geometrically correct the distortion caused by head movement in the laser ranging data and visual image data, thereby eliminating motion artifacts. The "hardware interrupt triggering mechanism" refers to the use of interrupt signals generated by a synchronous bus (such as I2C or SPI bus) based on the underlying hardware logic of FPGA (Field Programmable Gate Array) to forcibly write a unified timestamp at the physical moment of each frame of data acquisition from each sensor, which is different from the software-level time interpolation method. The "space-time data cube" refers to organizing multi-sensor fused data, after time alignment and spatial coordinate mapping, into a three-dimensional data structure that simultaneously includes spatial location, time, and multiple sensory information dimensions. The "environmental threshold comparison model" is a risk assessment model that pre-sets safety thresholds for various environmental parameters and compares the real-time acquired fused data with the preset thresholds. The "risk index" is a quantitative risk value obtained by weighting and calculating multiple indicators such as the degree of anomaly in the lining structure, gas concentration, and vibration intensity.

[0028] The multi-source approach in this solution includes at least three sensor data sources based on different physical principles, specifically laser ranging based on optical principles, visual images based on photoelectric imaging principles, motion attitude based on inertial principles, and gas detection based on electrochemical or infrared absorption principles. The term "high frequency" refers to the sampling frequency relative to the laser ranging sensor and the visual sensor; the IMU's sampling rate is typically several to tens of times higher to ensure that sufficiently dense motion attitude change information can be captured between two frames of laser or image data. The "millisecond-level time alignment" refers to the time synchronization error between sensor data frames being controlled within milliseconds. This precision is sufficient to ensure that, at normal walking speeds, the spatial mapping deviation of the multi-source data does not exceed millimeter levels. The "preset safety threshold" should be calibrated and set according to subway operation and maintenance industry standards and historical monitoring data of specific tunnel lines. The threshold for different lines and sections can be adaptively adjusted according to the actual engineering situation. The combination of visual, tactile, and auditory modalities emphasizes the simultaneous activation and synergistic effect of the three early warning channels, rather than the selective triggering of a single channel. The aim is to ensure that early warning information can be transmitted to workers through at least one effective channel under extreme perception conditions such as high noise and low light in tunnels.

[0029] In actual operation and maintenance of subway tunnels, inspection personnel need to monitor the condition of the tunnel lining while walking or even climbing. In existing technologies, wearable sensing devices inevitably sway during personnel movement, causing misalignment of laser point clouds and blurring of visual images. These motion artifacts render the collected data unusable for engineering reference. Furthermore, because laser sensors, visual sensors, and gas sensors have different sampling frequencies and communication protocols, relying solely on software-level timestamp interpolation for synchronization easily leads to spatiotemporal mismatches of data frames during personnel movement, causing "drift" in defect location. In addition, the tunnel environment is noisy and dimly lit, making a single buzzer alarm easily masked and unable to effectively notify personnel. This technical solution addresses these three major pain points by introducing a reverse geometric compensation mechanism in step S300 to solve the "inaccurate sampling" problem, hardware interruption time alignment and unified coordinate system mapping in step S400 to solve the "synchronization failure" problem, and localized risk calculation and tri-modal early warning combination in steps S500-S600 to solve the "interaction failure" problem. The core design logic of the entire monitoring method is as follows: first, high-quality raw data is obtained through physical vibration reduction and algorithm compensation; then, an accurate spatiotemporal data model is constructed through hardware-level synchronization; finally, risk assessment is completed at the local edge and early warning is output in a closed loop through multi-sensory channels. This forms a complete technical link from "precise perception" to "instant judgment" and then to "effective delivery". Moreover, the entire process does not rely on external network transmission, and the latency of the entire decision-making link is controlled at the millisecond level. It can still work autonomously even in areas deep in the tunnel without network coverage.

[0030] In one specific embodiment, after the worker enters the subway tunnel work area wearing the multi-source sensing fusion smart safety helmet, the smart safety helmet is powered on and activated, and each sensor module enters the working state. The laser ranging array 112 emits laser pulses at a set frequency to the tunnel lining surface and receives reflected signals, the wide dynamic range vision sensor 113 synchronously acquires wall images, and the gas sensor continuously monitors the concentration of harmful gases such as methane and hydrogen sulfide in the surrounding air. At the same time, the six-axis inertial measurement unit 111 captures the wearer's head posture change data in real time at a sampling rate several times higher than the above sensors. After receiving the data from each sensor, the edge computing unit inside the smart safety helmet first performs inverse geometric compensation on the laser and image data based on IMU data to eliminate motion artifacts caused by walking; secondly, it uses FPGA hardware interrupts to add a unified timestamp to each frame of data and complete coordinate system mapping to generate a fused space-time data cube; subsequently, the edge computing unit extracts the abnormal features of the lining structure, gas concentration values, and vibration intensity values ​​from the data cube and substitutes them into a weighted model to calculate the risk index. When the calculated risk index exceeds the preset safety threshold, the smart safety helmet brim light strip 131 immediately flashes red to provide a visual warning, the inner lining vibration device is activated simultaneously to apply tactile vibration to the wearer's head, and the bone conduction speaker 133 simultaneously outputs voice warning information. The three warning channels work together to ensure that the workers can perceive the danger signal and take avoidance measures as soon as possible.

[0031] In some embodiments, S300 specifically includes the following steps: S310, performing physical first-order vibration reduction through a flexible damping structure between the smart safety helmet liner and the sensor module; S320, acquiring the worker's attitude angular velocity and acceleration data in real time through a six-axis inertial measurement unit 111 at a sampling rate higher than that of the acquisition sensor; S330, based on the attitude angular velocity and acceleration data, performing inverse geometric compensation on the interval jitter frequency caused by the worker's step frequency in the laser ranging data and visual image data, and outputting stable data with motion artifacts removed.

[0032] The "flexible damping shock absorption structure" refers to the shock-absorbing connector installed between the inner lining of the smart helmet and the sensor module mounting base. It is made of a flexible material with elastic damping properties, passively absorbing and attenuating the mechanical vibrations transmitted to the sensor module by the wearer's head movements. The "first-order physical shock absorption" refers to the first-level (primary) vibration attenuation achieved through the elastic damping characteristics of the mechanical structure itself. This is a passive physical shock absorption system, forming a two-stage progressive anti-shake system with the algorithm compensation in subsequent steps. The "sampling rate higher than the acquisition sensor frequency" means that the IMU's data acquisition frequency is significantly higher than the frame rate of the laser rangefinder and vision sensor, ensuring that the IMU can capture multiple attitude sampling points within each frame of laser or image data acquisition time window, providing sufficient motion information for subsequent compensation calculations. The "attitude angular velocity" refers to the rotational angular rate about three orthogonal axes output by the gyroscope, used to describe the head's rotational motion state at the moment of acquisition. The "accelerometer data" refers to the linear acceleration values ​​along three orthogonal axes output by the accelerometer, used to describe the head's translational motion state. The "interval jitter frequency" refers to the periodic vibration frequency component transmitted to the head sensor caused by the gait rhythm during normal walking. This frequency typically falls within the range of 1Hz to 3Hz. The "motion artifact" refers to data distortion caused by sensor displacement or rotation during data acquisition. In laser ranging data, this manifests as point cloud misalignment or layering; in visual image data, it manifests as image trailing or blurring.

[0033] The physical first-order damping and subsequent inverse geometric compensation constitute a two-stage stabilization architecture of hardware passive damping + software active compensation. The role of physical first-order damping is to first attenuate large-amplitude mechanical shocks and high-frequency random vibrations, controlling the residual vibration transmitted to the sensor module within the range that the algorithm can compensate for. Without this physical damping step, the sensor module directly bears the entire amplitude of the head movement, and algorithm compensation alone will be insufficient to cover extreme amplitude situations. In the "sampling rate higher than the acquisition sensor frequency", the ratio between the IMU sampling rate and the laser / vision sensor frame rate should meet the engineering margin requirements of the Nyquist sampling condition, that is, the IMU sampling rate should be at least several times the upper limit of the target compensation frequency to ensure sufficient data density when the algorithm interpolates and reconstructs the motion trajectory. The inverse of the inverse geometric compensation means that the compensation direction is opposite to the actual motion direction - when the IMU detects that the sensor has undergone a certain angle of rotational offset or translational displacement at a certain acquisition moment, the algorithm corrects the frame data with a geometric transformation of equal magnitude and opposite direction, thereby mathematically restoring the data to the ideal result that should be obtained when the sensor is stationary.

[0034] Subway tunnel inspection personnel cannot keep their heads completely still during operations. The normal walking frequency is approximately 1.5Hz to 2.5Hz. This frequency, transmitted to the head through bones and muscles, generates periodic vibrations on the smart helmet, with the walking frequency as the fundamental frequency. If the sensor module is rigidly fixed directly to the helmet shell, this vibration will be fully transmitted to the sensor's optical and electronic components, causing periodic swaying of the laser emission direction during acquisition and periodic displacement of the visual sensor's photosensitive plane. Ultimately, this results in a jitter component with the same frequency as the walking frequency superimposed on the output data. The design concept of this solution is as follows: First, the flexible damping structure in step S310 achieves primary vibration attenuation at the physical level, converting most of the mechanical energy into heat dissipation through the deformation of the damping material, reducing the residual amplitude transmitted to the sensor to a small level. Based on this, step S320 uses a high-frequency IMU to accurately track the real-time attitude change trajectory of the residual vibration. Step S330 uses this trajectory data and a coordinate transformation algorithm to reverse-correct the residual offset in the laser and image data, ultimately outputting stable data with approximately no motion artifacts. This cascaded anti-shake solution combining "physics + algorithm" avoids the inherent limitations of purely mechanical shock absorption in completely eliminating low-frequency vibrations, and also avoids the risk of failure of purely algorithmic compensation when facing large-amplitude impacts. The two complement each other and achieve high-definition data output under dynamic walking conditions.

[0035] In some embodiments, S400 specifically includes the following steps: S410, at the bottom layer of the field-programmable gate array accelerator, a unified hardware timestamp is forcibly added to each frame of data acquired by the laser ranging sensor, vision sensor, inertial measurement unit, and gas sensor through an interrupt triggering method of the synchronization bus; S420, based on the hardware timestamp, millisecond-level time alignment is performed on sensor data with different sampling frequencies; S430, the aligned sensor data is mapped to a unified spatial coordinate system to generate a space-time data cube containing position, time, and multi-dimensional sensing information.

[0036] The "Field-Programmable Gate Array (FPGA) accelerator" is a hardware programmable logic device composed of numerous configurable logic units. It can be programmed using a hardware description language to implement specific digital logic functions, featuring strong parallel processing capabilities and high timing determinism, making it suitable for sensor synchronization control tasks with stringent time accuracy requirements. The "synchronization bus" refers to the physical communication link connecting each sensor to the FPGA. In this solution, an I2C bus or an SPI bus can be used, characterized by support for interrupt signal generation and transmission. The "interrupt triggering method" refers to the FPGA generating a hardware interrupt signal via the bus when it detects that a sensor has completed a frame of data acquisition. At the precise moment of the interrupt response, the current count value of the FPGA's internal high-precision clock is written into the timestamp field of that frame of data. This process is completed entirely at the hardware logic level, without software scheduling by the operating system, thus possessing deterministic nanosecond-level time accuracy. The "hardware timestamp" is a timestamp generated by a unified clock source within the FPGA at the moment of the hardware interrupt response. Compared to software timestamps, it is unaffected by uncertainties such as operating system scheduling delays and process switching. The "unified spatial coordinate system" refers to a three-dimensional Cartesian coordinate system established with a fixed reference point on the smart safety helmet as the origin. The output data of each sensor is transformed into this coordinate system through a calibrated extrinsic parameter matrix. The "space-time data cube" is a visual representation of the fused multidimensional data structure in this solution, which includes three dimensions: spatial dimension (three-dimensional coordinate position), temporal dimension (acquisition time), and sensory information dimension (laser distance value, image features, gas concentration, vibration intensity, and other physical quantities).

[0037] The interrupt triggering method described above differs fundamentally from conventional software polling timestamp schemes. In software polling, the processor sequentially queries the data readiness status of each sensor, reads the system clock, and adds a timestamp only after receiving a readiness signal. The order of queries and the jitter of operating system interrupt responses introduce uncertain time delays. In contrast, the hardware interrupt triggering method used in this scheme allows the FPGA's parallel logic circuits to directly respond to the data readiness hardware signals of each sensor, timestamping each data stream within the same clock domain. The timestamp generation processes of each sensor are independent and do not interfere with each other, thus ensuring millisecond-level or even sub-millisecond-level synchronization accuracy. In practical engineering, millisecond-level time alignment means that when a person walks at a normal pace, data collected by different sensors at the same physical moment can precisely correspond on the timeline, and the spatial mapping deviation caused by synchronization errors is far smaller than the smallest scale of the disease feature of interest. The establishment of the unified spatial coordinate system requires the prior acquisition of the rotation matrix and translation vector of each sensor relative to the reference coordinate system of the helmet body 1 through a calibration program (i.e., external parameter calibration). This calibration is performed before the smart helmet leaves the factory or after the sensor module is installed.

[0038] In subway tunnel operation and maintenance scenarios, it is necessary to simultaneously collect data on multiple physical quantities for comprehensive judgment. However, due to differences in physical principles and hardware characteristics, the sampling frequencies of various sensors differ significantly—the scanning frequency of laser rangefinders, the frame rate of vision sensors, the output frequency of IMUs, and the response period of gas sensors all vary. When personnel are moving, if the time synchronization accuracy between the sensor data is insufficient, even if each data stream is individually of acceptable quality, spatial fusion will result in misaligned mapping positions due to time mismatch. For example, if a wall point marked by laser rangefinder data has a time deviation of tens of milliseconds from the visual image data, the sensor will have undergone considerable spatial displacement during pedestrian movement. When the two are superimposed in the same coordinate system, the corresponding defect feature points will show significant drift, severely affecting the accuracy of crack location and defect quantification. This solution implements hardware-level time synchronization at the FPGA level, fundamentally eliminating the time uncertainty caused by software scheduling at the physical level, ensuring strict alignment of heterogeneous sensor data on the time axis. Based on this, the data from each sensor are further transformed into the same spatial coordinate system through a pre-calibrated extrinsic parameter matrix, so that the perception information from different sensors can be accurately superimposed on the same spatial location in a physical sense. Finally, a data cube containing complete spatiotemporal information and multi-dimensional perception information is constructed, providing a high-fidelity data foundation for subsequent anomaly detection and risk calculation.

[0039] In S500, the formula for calculating the risk index is as follows: R=W1×ΔD + W2×G +W3×V; Where R is the risk coefficient, Δ D , where is the abrupt change in laser ranging, characterizing the degree of structural anomaly of cracks or voids on the lining surface; G is the ambient gas concentration value; V is the ambient vibration intensity value; W1, W2, and W3 are the weighting coefficients corresponding to each index.

[0040] The "risk index R" is a quantitative representation of the overall safety risk level within the current tunnel section. It is a dimensionless comprehensive score value; the higher the value, the higher the risk level of the current environment. The "laser ranging mutation Δ" D "Ambient gas concentration value G" refers to a significant jump in the laser ranging return value between adjacent scan frames or adjacent scan points. This jump deviates from the continuous and gradual change pattern of the normal lining surface, usually indicating structural anomalies such as cracks, voids, and spalling on the lining surface. "Ambient gas concentration value G" refers to the real-time output value of the concentration of harmful gases detected by the gas sensor. In the subway tunnel environment, harmful gases that need to be monitored include methane, hydrogen sulfide, and carbon monoxide, as exceeding the concentration standard will directly threaten the life safety of workers. "Ambient vibration intensity value V" refers to the intensity of tunnel environmental vibration detected by the inertial measurement unit or dedicated vibration sensor. This vibration may originate from the passage of trains on adjacent lines, changes in tunnel surrounding rock stress, or the operation of construction machinery. Abnormal vibration may indicate changes in structural bearing capacity or external interference risks. "Weighting coefficients W1, W2, and W3" are weight values ​​set according to the degree of influence of each risk factor on tunnel operation safety, used to reflect the relative importance of different risk indicators in the comprehensive assessment.

[0041] The safety risks faced in tunnel operation and maintenance are not singular, but rather a comprehensive result of the combined effects of multiple factors, including structural health, environmental gas conditions, and external vibration. Focusing solely on structural cracks while ignoring gas leaks, or focusing only on gas concentration while neglecting the risk of structural collapse, may lead to the overlooking of significant safety hazards. This solution addresses these hazards by addressing abnormalities in the lining structure (Δ...). D Real-time detection data from three dimensions—gas concentration (G), vibration intensity (V)—are incorporated into a unified weighted calculation model, enabling a comprehensive risk assessment of the tunnel working environment. This model runs in real-time within a local edge computing unit, outputting a risk index R value each calculation cycle. An early warning process is triggered when the R value exceeds a preset safety threshold. Since all calculations are performed locally on the smart safety helmet, without relying on external network connections, risk assessment and early warning decisions can still be executed normally even in environments with no signal coverage deep within the tunnel, with response latency controlled to the millisecond level.

[0042] In some embodiments, in S500, the determination of abnormal information on the lining surface structure includes the following judgment conditions: when the laser reflectivity of the lining surface is abnormal or the ranging value changes abruptly, it is determined that there is a risk of cracks or voids; when the gas concentration exceeds the preset safety boundary value, it is determined that there is a risk of harmful gas exceeding the standard.

[0043] It should be understood that the "abnormal laser reflectivity of the lining surface" refers to the intensity of the reflected signal returned after a laser pulse irradiates the tunnel lining surface deviating from the statistical average range of the normal reflectivity of that section of lining material. On a healthy concrete lining surface, laser reflectivity exhibits relatively stable statistical characteristics; however, when defects such as water seepage, efflorescence, or exposed reinforcement appear on the lining surface, changes in surface material and roughness lead to abnormal fluctuations in reflectivity. The "abrupt change in distance measurement value" refers to a step change in the distance returned by the laser ranging sensor in the same spatial area between adjacent scanning points or frames during continuous scanning, exceeding the normal gradual change characteristics of the curved surface. This abrupt change usually corresponds to cracks (surface depressions or openings causing a sudden increase in distance) or voids (vacuuming behind the lining causing abnormal echo characteristics in a specific area) on the lining surface. The "preset safety boundary value" refers to the concentration limits set for various harmful gases in the tunnel according to relevant national safety standards and subway operation and maintenance industry specifications. When the real-time detected concentration reaches or exceeds this limit, the environment is deemed not to meet operational safety conditions.

[0044] In the actual operation and maintenance of subway tunnels, the health of the lining structure and the safety of ambient gases are two core factors affecting the safety of workers. Over long-term operation, tunnel linings are affected by groundwater erosion, ground subsidence, train vibration, and other factors, which may gradually lead to defects such as cracks, leaks, and voids. If not detected in time, these can develop into major hidden dangers endangering train operation and personnel safety. Simultaneously, subway tunnels, especially those under construction or undergoing renovation, may pose a risk of accumulating harmful gases such as methane seeping from the ground and carbon monoxide generated during construction. This solution incorporates both structural anomaly identification and gas concentration identification into the risk assessment system, ensuring comprehensive monitoring of the two main threat sources in the tunnel working environment. Laser sensors can detect macroscopic structural deformation through the spatial gradient of distance measurements and identify microscopic surface degradation through reflectivity characteristics. The synergistic effect of these two indicators improves both the sensitivity and reliability of defect detection.

[0045] In some embodiments, S600 specifically includes the following steps: S610, when the risk index exceeds a preset safety threshold, visual warning is given by the flexible light strip 131 on the brim of the smart safety helmet flashing red; and, tactile vibration is applied to the worker's head by the vibration feedback device 132 that fits into the inner lining of the smart safety helmet; and, voice warning information is transmitted to the worker by the bone conduction speaker 133.

[0046] The three warning modalities—visual (red flashing light strip 131), tactile (lining vibration), and auditory (bone conduction speech)—are activated simultaneously and output in parallel when a warning is triggered, rather than being triggered sequentially or selectively. The core consideration for this design lies in the unique working environment of subway tunnels: the tunnels are extremely dark, but workers wear headlamps and their attention is usually focused on the work surface, so a single visual warning may not be within their field of vision; the noise from passing trains in the tunnels is extremely high, and a single auditory warning may be masked by the ambient noise; while the tactile channel acts directly on the body and is unaffected by light and noise, but a single tactile stimulus, if of limited intensity, may be confused with the vibrations during work. The simultaneous output of the three modalities creates redundant coverage, ensuring that at least one channel can effectively convey warning information under any extreme working conditions.

[0047] In the unique, enclosed space of a subway tunnel, traditional single-ring alarm systems have significant limitations. The high-intensity background noise generated by the mixture of construction machinery noise, train sounds, and pipeline airflow can easily drown out conventional buzzer sounds. Simultaneously, the extremely poor lighting conditions inside tunnels mean that conventional indicator light signals are easily overlooked when workers are highly focused on their tasks. This multimodal early warning design is based on a thorough analysis of the extreme sensory environment of tunnels: a red flashing light strip 131 around the brim of the helmet covers the worker's surrounding visual perception area; an inner lining vibration device directly stimulates the tactile nerves in the head, making it impossible to ignore; and a bone conduction speaker 133 bypasses the airborne propagation path of environmental noise to directly transmit the voice signal to the inner ear. These three channels complement and redundancy each other, forming a closed-loop delivery of warning information—regardless of the worker's current sensory state (looking ahead, facing away from the light, at the peak of noise, etc.), the warning information is effectively delivered, maximizing the safety of frontline workers.

[0048] The smart safety helmet integrates an analysis and control unit, and the anomaly detection, risk assessment, and early warning decision-making in S500 and S600 are all completed under the analysis and control of the analysis and control unit of the smart safety helmet.

[0049] This invention also proposes an intelligent safety helmet for implementing a safety monitoring method for subway tunnel operations. The specific details of this method are as described in the above embodiments. The intelligent safety helmet includes a helmet body 1, a sensing unit 11, a processing unit 12, and an interaction unit 13. The sensing unit 11 is located at the front end of the brim of the helmet body 1 and includes a six-axis inertial measurement unit 111, a laser ranging array 112, a wide dynamic range vision sensor 113, and an acoustic environment monitoring unit 114. The processing unit 12 is embedded inside the helmet body 1 and includes a low-power edge computing board 121. This board integrates a programmable gate array accelerator for performing sensor data fusion, pose compensation, and risk assessment. The interaction unit 13 includes a light strip 131 disposed around the helmet body 1, a vibration feedback device 132 fitted into the inner lining of the helmet body 1, and a bone conduction speaker 133 for outputting multimodal warning information to workers.

[0050] The "helmet body 1" refers to the shell structure conforming to industrial smart safety helmet standards, including the outer shell, inner liner, brim, and wearing adjustment components, providing basic physical head protection and a mounting carrier for each functional unit of the entire smart safety helmet. The "sensing unit 11" refers to a collection of sensor modules responsible for collecting various physical information about the tunnel environment. It is positioned at the front of the brim to ensure that the sensor's detection field of view faces the wearer's direction of travel and gaze, i.e., the tunnel wall and the space in front. The "laser ranging array 112" refers to a sensor group composed of multiple laser ranging sensors arranged in a specific spatial pattern. Working in array form expands the coverage of a single scan and improves the efficiency of collecting wall structure information. The "wide dynamic range vision sensor 113" refers to a camera sensor with high dynamic range imaging capabilities, capable of maintaining clear details in dark areas and preventing overexposure in bright areas under complex lighting conditions in tunnels with extremely low illumination and strong reflective areas (such as reflective signs and water reflections). The "acoustic environment monitoring unit 114" refers to a microphone or acoustic sensor used to collect acoustic information of the tunnel environment, which can be used to detect abnormal sounds (such as structural cracking sounds, gas leak hissing sounds, etc.). The "processing unit 12" refers to the core module responsible for data processing and logic control, embedded inside the safety helmet body 1 for compact integration. The "low-power edge computing board 121" refers to an embedded computing platform with edge AI inference and real-time signal processing capabilities under limited power consumption budget. The "interaction unit 13" refers to a set of functional modules responsible for outputting system decision results to the wearer in a human-perceptible form.

[0051] The overall structural design of this smart safety helmet follows an architecture concept of three-layer separation and physical integration of perception, computing, and interaction. The technical rationale for this design lies in: clearly layering functions logically, facilitating independent development, testing, maintenance, and upgrades of each unit; and ensuring that hardware iteration of one unit does not affect the normal operation of other units. Simultaneously, physically integrating the three units within the same smart safety helmet avoids the burden of the wearer carrying additional external devices or connecting cables, achieving the design goal of individual operation. Compared to existing basic smart safety helmets that only have positioning functions, this smart safety helmet integrates a multi-source sensor array, edge AI computing power, and a multimodal interaction device within the same shell space. This brings the tunnel wall scanning and inspection, environmental safety monitoring, and real-time early warning functions—traditionally requiring large inspection vehicles or specialized instruments—down to the level of personal wearable devices. This allows ordinary inspection personnel to complete large-area tunnel structural health monitoring during daily patrols, transforming the subway operation and maintenance mode from specialized large-scale inspections to routine individual self-inspections. Meanwhile, because the processing unit 12 has independent edge computing capabilities, the smart safety helmet can complete all data analysis and safety decisions without relying on external networks and servers. It can still work independently in areas deep in tunnels without signal coverage, ensuring the all-weather availability of safety protection functions.

[0052] In some embodiments, such as Figure 2 As shown, the inner lining of the safety helmet body 1 is provided with a shock-absorbing module 14. The shock-absorbing module 14 includes a flexible damping bracket made of silicone shock-absorbing material, which is used to absorb the vibration generated by the movement of the worker's head and realize physical vibration isolation of the sensor module.

[0053] The design of the vibration damping module 14 is a structural implementation of the physical first-order vibration damping function in the aforementioned method step S310. This module is positioned between the inner lining of the smart safety helmet and the sensor module. Its structural design needs to simultaneously meet two engineering requirements: first, vibration damping performance, meaning the stiffness and damping parameters of the flexible damping bracket should match the vibration attenuation requirements within the human walking frequency range, ensuring that the residual vibration amplitude transmitted to the sensor module is reduced to a range that can be handled by subsequent algorithm compensation; second, structural support stability, meaning that while providing elastic vibration isolation, the flexible damping bracket must still provide reliable spatial positioning and load-bearing support for the sensor module, preventing uncontrollable displacement of the sensor installation position due to excessive deformation.

[0054] During actual use of smart safety helmets, the worker's walking, turning, bending, and climbing movements inevitably generate multi-directional vibrations transmitted to the head. If the sensor module (especially the laser ranging array 112 and the vision sensor) is rigidly connected to the smart safety helmet shell, these vibrations will be directly transmitted to the optical and photosensitive elements of the sensor without attenuation. This causes the laser emission direction to deflect at the moment of acquisition and the photosensitive surface of the camera to shift, thus superimposing motion artifacts into the output data. This solution solves this problem by setting a shock-absorbing module 14 between the sensor module and the inner lining of the smart safety helmet. The elastic deformation of the silicone flexible damping bracket converts the vibration kinetic energy into heat energy dissipated by internal friction of the material, forming a physical barrier in the vibration transmission path. This passive physical damping consumes no electrical energy, requires no control logic, and has no response delay. It can immediately attenuate vibrations at the moment they occur, laying a good data foundation for subsequent IMU reverse geometric compensation algorithms. After physical damping, the residual vibration amplitude and frequency range experienced by the sensor are significantly narrowed, and subsequent algorithms only need to handle small-amplitude fine compensation, ensuring both computational accuracy and stability.

[0055] In some embodiments, the processing unit 12 and the sensing unit 11 are respectively located on the front and rear sides of the helmet body 1.

[0056] The split layout design, with the sensing unit 11 located at the front and the processing unit 12 at the rear, is not an arbitrary spatial arrangement, but rather a comprehensive optimization result based on two engineering constraints. The first constraint is functional: the laser ranging array 112 and the vision sensor in the sensing unit 11 need to be oriented towards the wearer's direction of travel to obtain detection data from the tunnel walls in front and to the sides, therefore they must be positioned at the front of the brim; while the processing unit 12 has no specific orientation requirements and can be flexibly arranged in other areas of the helmet. The second constraint is ergonomic: the sensor modules in the sensing unit 11 (especially the optical components of the laser ranging array 112 and the vision sensor) have a certain weight. If the weight of the smart safety helmet is too concentrated at the front, it will cause the helmet to tilt forward, requiring the wearer to continuously tighten the muscles at the back of their neck to maintain balance, leading to neck fatigue and discomfort after prolonged use. By placing a processing unit 12 (edge ​​computing board and matching battery module) with a certain weight at the rear of the helmet, a counterweight balance effect can be formed in the front-to-back direction of the helmet, so that the center of gravity of the smart safety helmet is close to the natural balance point of the wearer's head, thereby significantly improving wearing comfort.

[0057] In terms of specific structure, the smart safety helmet also includes an adjustable strap 15 for wearing and fixing the safety helmet body 1, and a battery structure 16 that can provide power to the monitoring unit, processing unit 12, etc.

[0058] The above description is merely an exemplary embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention specification and drawings under the technical concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A safety monitoring method for subway tunnel operations, characterized in that, The monitoring steps include the following: S100, a smart safety helmet that acquires multi-source sensing fusion; S200: The multi-source sensor array integrated on the smart safety helmet collects multi-source data information in the tunnel environment, including laser ranging data, visual image data, and ambient gas data. S300: The six-axis inertial measurement unit on the smart safety helmet collects high-frequency motion trajectory data of the wearer during walking, and performs inverse geometric compensation on the laser ranging data and visual image data using the high-frequency motion trajectory data; S400, based on the hardware interrupt triggering mechanism in the smart safety helmet, performs millisecond-level time alignment on sensor data with different sampling frequencies and performs spatial mapping in a unified coordinate system to construct a fused space-time data cube; S500. Based on the space-time data cube, establish an environmental threshold comparison model, and perform weighted calculations by comprehensively considering the abnormal information of the lining surface structure, gas concentration information, and vibration intensity information to generate a risk index. S600. When the risk index exceeds the preset safety threshold, a warning is issued to the operator through a combination of visual, tactile and auditory modalities.

2. The safety monitoring method for subway tunnel operations as described in claim 1, characterized in that, S300 specifically includes the following steps: S310, physical first-order shock absorption is achieved through the flexible damping shock absorption structure between the inner lining of the smart safety helmet and the sensor module. S320: The six-axis inertial measurement unit acquires the operator's attitude angular velocity and acceleration data in real time at a sampling rate higher than that of the acquisition sensor. S330. Based on the attitude angular velocity and acceleration data, perform inverse geometric compensation on the interval jitter frequency caused by the person's walking frequency in the laser ranging data and visual image data, and output stable data to remove motion artifacts.

3. The safety monitoring method for subway tunnel operations as described in claim 1, characterized in that, The S400 specifically includes the following steps: S410, at the bottom layer of the field programmable gate array accelerator, uses the interrupt triggering method of the synchronous bus to force a unified hardware timestamp on each frame of data acquired by the laser rangefinder, vision sensor, inertial measurement unit and gas sensor. S420. Based on the hardware timestamp, perform millisecond-level time alignment on sensor data with different sampling frequencies; S430: Map the aligned sensor data to a unified spatial coordinate system to generate a space-time data cube containing location, time, and multi-dimensional sensing information.

4. The safety monitoring method for subway tunnel operations as described in claim 1, characterized in that, In S500, the formula for calculating the risk index is as follows: R=W1×Δ D + W2×G +W3×V; Where R is the risk coefficient, Δ D , where is the abrupt change in laser ranging, characterizing the degree of structural anomaly of cracks or voids on the lining surface; G is the ambient gas concentration value; V is the ambient vibration intensity value; W1, W2, and W3 are the weighting coefficients corresponding to each index.

5. The safety monitoring method for subway tunnel operations as described in claim 2, characterized in that, In S500, the determination of abnormal information on the lining surface structure includes the following judgment conditions: When the laser reflectivity of the lining surface is abnormal or the ranging value changes abruptly, it is determined that there is a risk of cracks or voids. When the gas concentration exceeds the preset safety boundary value, it is determined that there is a risk of excessive harmful gas.

6. The safety monitoring method for subway tunnel operations as described in claim 2, characterized in that, The S600 specifically includes the following steps: S610. When the risk index exceeds the preset safety threshold, a visual warning is given by flashing red light strip on the brim of the smart safety helmet; and, a tactile vibration is applied to the worker's head by a vibration feedback device that fits into the inner lining of the smart safety helmet; and, a voice warning is transmitted to the worker by a bone conduction speaker.

7. The safety monitoring method for subway tunnel operations as described in claim 1, characterized in that, The smart safety helmet integrates an analysis and control unit. The anomaly detection, risk assessment, and early warning decision-making in S500 and S600 are all completed under the analysis and control of the analysis and control unit of the smart safety helmet.

8. A smart safety helmet for implementing the safety monitoring method for subway tunnel operations according to any one of claims 1-7, characterized in that, The smart safety helmet includes: The helmet itself; The sensing unit is located at the front end of the brim of the safety helmet body, and includes a six-axis inertial measurement unit, a laser ranging array, a wide dynamic range vision sensor, and an acoustic environment monitoring unit. The processing unit, embedded inside the helmet body, includes a low-power edge computing board. This board integrates a programmable gate array accelerator for performing sensor data fusion, pose compensation, and risk assessment. The interactive unit includes a light strip disposed around the periphery of the safety helmet body, a vibration feedback device fitted into the inner lining of the safety helmet body, and a bone conduction speaker, used to output multimodal warning information to the worker.

9. The smart safety helmet as described in claim 8, characterized in that, The safety helmet body has a shock-absorbing module on its inner lining. The shock-absorbing module includes a flexible damping bracket made of silicone shock-absorbing material, which is used to absorb the vibration generated by the movement of the worker's head and achieve physical vibration isolation of the sensor module.

10. The smart safety helmet as described in claim 8, characterized in that, The processing unit and the sensing unit are respectively located on the front and rear sides of the helmet body.

Images