Tunnel multi-source monitoring system with edge intelligence and autonomous regulation and control capability

Abstract

The invention relates to the technical field of tunnel structure safety monitoring and intelligent control, in particular to a tunnel multi-source monitoring system with edge intelligence and autonomous regulation and control capabilities, which comprises a multi-source sensor acquisition module, an edge calculation processing module, a state sensing and adaptive regulation and control module and a local early warning and execution module, the abnormity self-diagnosis module is used for monitoring the operation state of each module of the system in real time and performing isolation, switching or data compensation when a fault occurs; and the communication and remote interaction module is used for carrying out data interaction with a remote platform through wired and wireless multilinks and executing breakpoint resume and data integrity verification after communication is recovered. According to the invention, data on-site processing, state sensing self-adaptive regulation and control, abnormity self-diagnosis and local early warning are realized, safe operation can still be ensured under the condition that communication is limited or interrupted, and a remote platform is synchronized during recovery.

Description

Technical Field

[0001] This invention relates to the field of tunnel structure safety monitoring and intelligent control technology, and in particular to a tunnel multi-source monitoring system with edge intelligence and autonomous adjustment capabilities, which is suitable for complex tunnel operating environments characterized by multi-source heterogeneous sensor deployment, inconsistent data sampling frequencies, and drastic changes in operating conditions. Background Technology

[0002] With the rapid development of transportation infrastructure construction, a large number of tunnel projects have been put into operation under complex geological and variable environmental conditions. During their service, tunnels may face a variety of dynamic conditions, such as seismic impact, sudden water inrush, surrounding rock instability, overload impact, and construction disturbance. These conditions are often accompanied by sudden and frequent fluctuations in multi-source monitoring data, including strain, displacement, seepage pressure, temperature and humidity, acoustic emission signals, and video images. These fluctuations are rapid and short-lived, placing higher demands on the response speed and data processing capabilities of the monitoring system.

[0003] Existing tunnel monitoring systems mostly rely on a centralized cloud processing architecture, where front-end sensor nodes transmit raw monitoring data in real-time or near real-time to a remote data center for unified processing and analysis. While this centralized architecture offers concentrated computing resources and convenient algorithm updates, it suffers from the following shortcomings in practical applications:

[0004] 1. Significant transmission latency, failing to meet millisecond-level response requirements. In emergencies, data collected by monitoring nodes needs to be transmitted to the cloud via wireless or wired links. Due to limitations in bandwidth, network congestion, and link quality, there is often a delay of hundreds of milliseconds to several seconds. This can lead to missed opportunities for optimal response in rapidly changing disaster conditions.

[0005] 2. Difficulty in fusing data from multi-source heterogeneous sensors. Tunnel monitoring involves a variety of sensor types, including strain gauges, displacement gauges, temperature and humidity sensors, fiber optic sensors, cameras, and acoustic emission probes. These sensors have different sampling frequencies, ranges and accuracies, and asynchronous acquisition clocks, directly leading to misalignment of the data time axis and information mismatch, thus affecting the accuracy and reliability of multimodal fusion analysis.

[0006] 3. Lack of edge-side state awareness and adaptive control capabilities. Most existing systems only have data acquisition and uploading functions, and cannot dynamically adjust the sampling frequency, transmission mode, and alarm threshold according to changes in operating conditions, making it difficult to achieve rapid response through local decision-making in the early stages of a disaster.

[0007] 4. Insufficient security during communication disruptions. In the complex environment of tunnels, communication links may be interrupted due to equipment failure, power supply anomalies, strong electromagnetic interference, or structural damage. Once the cloud and front-end lose connection, the existing system cannot continue to operate effective safety monitoring and control measures, which may lead to the expansion of disasters or delays in response.

[0008] 5. Lack of local self-diagnostic capabilities. After long-term operation, some monitoring nodes may experience problems such as sensor drift, data packet loss, and abnormal power consumption. Lacking real-time self-diagnostic and fault isolation mechanisms, they cannot promptly restore monitoring capabilities or issue alarms to maintenance personnel.

[0009] In summary, existing tunnel monitoring systems suffer from drawbacks such as long response delays, difficulties in data fusion, lack of edge intelligent control, and insufficient local emergency response capabilities when dealing with high-frequency data scenarios under dynamic operating conditions. Therefore, there is an urgent need for a tunnel monitoring system and method capable of performing multi-source heterogeneous data preprocessing, sampling synchronization, state awareness, and adaptive control at the edge, and possessing local early warning, anomaly self-diagnosis, and communication recovery capabilities to ensure the safety and continuity of tunnel operation. Summary of the Invention

[0010] Based on the above description, the present invention provides a tunnel multi-source monitoring system with edge intelligence and autonomous control capabilities, which realizes local data processing, state-aware adaptive control, anomaly self-diagnosis and local early warning, ensuring safe operation even when communication is restricted or interrupted and synchronizing with the remote platform when recovery is achieved.

[0011] On the one hand, the technical solution of the present invention to solve the above-mentioned technical problems is as follows: a tunnel multi-source monitoring system with edge intelligence and autonomous control capabilities, comprising:

[0012] The multi-source sensor acquisition module is used to acquire multimodal parameters of the tunnel structure and environment, and attach high-precision time tags at the moment of acquisition;

[0013] The edge computing processing module is used to perform real-time preprocessing, sampling synchronization, and multi-source fusion of the raw data from the multi-source sensor acquisition module at the tunnel site.

[0014] The status awareness and adaptive control module is used to identify the tunnel operating status in real time and dynamically adjust the sampling frequency, power consumption mode, alarm threshold and data upload frequency based on the operating condition feature vector output by the edge computing processing module.

[0015] The local early warning and execution module is used to directly trigger an early warning and link the execution mechanism at the tunnel site when the risk level output by the state perception and adaptive control module reaches a preset threshold.

[0016] The anomaly self-diagnosis module is used to monitor the operating status of each module in the system in real time and to isolate, switch or compensate data when a fault occurs.

[0017] The communication and remote interaction module is used to interact with the remote platform via wired and wireless multi-links, and to perform breakpoint resume and data integrity verification after communication is restored.

[0018] This invention achieves high-precision synchronous acquisition of multi-modal parameters of tunnel structure and environment through a multi-source sensor acquisition module, ensuring consistency of different types of sensor data in the time dimension and providing a reliable foundation for subsequent data fusion. An edge computing processing module preprocesses, synchronizes sampling, and fuses the raw data on-site, reducing data transmission volume and improving real-time analysis capabilities, effectively reducing cloud processing latency. Based on the fused operating condition feature vector, the state perception and adaptive control module can determine the tunnel's operating status in real time and dynamically adjust the sampling frequency, power consumption mode, alarm threshold, and data reporting frequency, enabling on-demand optimization of monitoring strategies and improving system performance. The system exhibits adaptability and energy efficiency under the same operating conditions. The local early warning and execution module can directly trigger on-site warnings via sound, light, display, and voice signals when communication is restricted or interrupted, and coordinate with ventilation, lighting, traffic guidance, and drainage systems to ensure rapid response to emergencies. The anomaly self-diagnosis module performs real-time health monitoring and fault detection on sensors, computing units, and communication links, performing isolation, switching, or data compensation upon detection of anomalies to maintain continuous system operation. The communication and remote interaction module ensures stable connection with the remote platform through multiple links, and performs breakpoint resume transmission and data integrity verification after link restoration, ensuring complete historical data transmission and synchronous updates. Therefore, this invention realizes a closed-loop intelligent monitoring system integrating multi-source synchronous acquisition, edge intelligent processing, adaptive control, local rapid response, anomaly self-diagnosis, and robust communication, significantly improving the tunnel's real-time perception, rapid response, and safety assurance capabilities under dynamic operating conditions.

[0019] Based on the above technical solution, the present invention can be further improved as follows.

[0020] Furthermore, the multi-source sensor acquisition module includes strain sensors, displacement sensors, cross-sectional convergence monitoring sensors, seepage pressure sensors, temperature and humidity sensors, acoustic emission sensors, and video image sensors. Each sensing node has a built-in local cache and time compensation unit, and achieves millisecond-level or sub-millisecond-level time alignment through IEEE 1588 Precision Time Protocol (PTP), GPS timing, or hardware pulse synchronization signals.

[0021] Through the above technical solution, strain sensors, displacement sensors, cross-sectional convergence monitoring sensors, seepage pressure sensors, temperature and humidity sensors, acoustic emission sensors, and video image sensors are configured in the multi-source sensor acquisition module, realizing comprehensive and multi-modal perception of tunnel structure deformation, surrounding rock pressure, environmental parameters, and structural damage characteristics. Each sensor node has a built-in local cache and time compensation unit, which can temporarily store data at the moment of acquisition and correct time errors caused by sensor response delay or transmission delay, effectively avoiding data loss and timing deviation. By adopting the IEEE 1588 Precision Time Protocol (PTP), GPS timing, or hardware pulse synchronization signal, millisecond-level or sub-millisecond-level time alignment of sampling from different types of sensors is achieved, providing a high-precision timing foundation for subsequent accurate fusion and consistency analysis of multi-source data, thereby significantly improving the accuracy and reliability of working condition identification and risk assessment.

[0022] Furthermore, the edge computing processing module adopts a heterogeneous architecture of ARM processor and FPGA or embedded GPU, runs preprocessing algorithms including Kalman filtering, wavelet denoising and outlier removal, uses interpolation, resampling and time window alignment to achieve sampling synchronization, and generates working condition feature vectors based on weighted multimodal fusion or LSTM combined with attention mechanism model.

[0023] Through the above technical solutions, a heterogeneous architecture of ARM processor and FPGA or embedded GPU is adopted in the edge computing processing module, realizing the coordinated operation of general processing and parallel acceleration. This ensures the flexibility of the algorithm and significantly improves the real-time performance of large-scale data processing. Running preprocessing algorithms such as Kalman filtering, wavelet denoising, and outlier removal can effectively suppress noise, remove outliers, and improve the quality and stability of the original data. Interpolation, resampling, and time window alignment are used to achieve sampling synchronization of data from different types of sensors, eliminating timing misalignments caused by differences in sampling frequency and clock. On this basis, a working condition feature vector is generated based on a weighted multimodal fusion model or a deep learning model combining LSTM and attention mechanisms, realizing deep fusion and temporal feature extraction of multi-source heterogeneous data, thereby improving the accuracy and real-time performance of working condition identification and risk assessment under dynamic working conditions.

[0024] Furthermore, the state perception and adaptive control module calculates the risk level using the following formula:

[0025] ;

[0026] in, For strain standard deviation, For the standard deviation of displacement, For acoustic emission event rate, For the rate of change of the image, The weighting coefficients are obtained from training with historical data; when the risk level is ≥3, the sampling frequency is increased to twice the original value and the alarm trigger threshold is lowered.

[0027] By employing the aforementioned technical solution, multi-source features such as strain standard deviation, displacement standard deviation, acoustic emission event rate, and image change rate are weighted and fused to comprehensively reflect the overall safety status of the tunnel structure and environment. Each weight coefficient is obtained through training with historical data, making the calculation model targeted and adaptive. When the risk level reaches or exceeds level 3, the system automatically increases the sampling frequency to twice the original value and lowers the alarm trigger threshold, achieving rapid response and high-precision monitoring of high-risk conditions. This mechanism can significantly improve the spatiotemporal resolution and alarm sensitivity of monitoring in the early stages of increased risk, thereby shortening the response time to emergencies, reducing accident risks, and improving the safety and reliability of tunnel operation.

[0028] Furthermore, the local early warning and execution module includes an audible and visual alarm, an electronic display screen, a voice broadcasting unit, and an execution control unit connected to a ventilation system, a lighting system, a traffic guidance device, and a barrier gate. The execution control unit communicates with the actuator via a CAN bus or an RS485 bus and completes the action within 100 ms after receiving a trigger signal.

[0029] Furthermore, the abnormal self-diagnosis module determines faults by detecting sensor drift, system crash, data loss, communication link abnormalities, and processing unit overload, calculates the health index HI, triggers a maintenance alarm when HI < 0.7, and calls redundant sensor data or data based on model prediction for compensation when some sensors fail.

[0030] The communication and remote interaction module includes industrial Ethernet, 5G communication, LoRa and Wi-Fi Mesh communication units. It has a multi-link automatic switching function, switches to the backup link when the main link is interrupted, and performs breakpoint resume transmission based on the timestamp index after the link is restored. It uses CRC or MD5 check to ensure data integrity and supports local storage and seamless synchronization of at least 1 hour of continuous data.

[0031] Secondly, the technical solution of the present invention to solve the above-mentioned technical problems is as follows: a tunnel multi-source monitoring method with edge intelligence and autonomous control capabilities, comprising:

[0032] Collect multi-source sensor data within the tunnel monitoring area and synchronize the data in time.

[0033] At the edge, preprocessing, synchronization, and multi-source fusion of the collected data are performed to generate characteristic data representing the working conditions.

[0034] Calculate the risk level based on the aforementioned feature data;

[0035] The sampling frequency, power consumption mode, alarm threshold, and data reporting frequency of the monitoring system are dynamically adjusted according to the risk level.

[0036] When the risk level reaches the preset conditions, an early warning is triggered at the tunnel site and the execution mechanism is activated.

[0037] The system performs anomaly detection during operation and isolates, switches, or compensates for data when a fault occurs.

[0038] When the communication link is restored, perform breakpoint resume transmission, integrity verification and remote synchronization of data.

[0039] Through the above technical solutions, multi-source sensor data is collected in the tunnel monitoring area and high-precision time synchronization is performed to ensure the consistency of different types of data in terms of time sequence; preprocessing, synchronization, and multi-source fusion are performed at the edge end, which significantly reduces data transmission latency and improves real-time analysis capabilities; risk level is calculated based on fused feature data, and sampling frequency, power consumption mode, alarm threshold, and data reporting frequency are dynamically adjusted according to the risk level to achieve adaptive optimization of the monitoring strategy; when the risk level reaches the preset conditions, an early warning is immediately triggered on site and the actuator is linked to ensure rapid handling of emergencies; anomaly detection is implemented during operation, and isolation, switching, or data compensation is performed when a fault occurs to ensure monitoring continuity and data integrity; after communication is restored, breakpoint resume transmission and integrity verification are performed to ensure complete backhaul and remote synchronization of historical data, thereby achieving efficient perception, intelligent control, and safety assurance of tunnel operation status under dynamic working conditions.

[0040] Furthermore, the risk level calculation steps include:

[0041] Based on strain standard deviation standard deviation of displacement Acoustic emission event rate and image change rate Calculate risk level value The calculation formula is as follows: ;in, These are the weight coefficients obtained by training based on historical operating data.

[0042] Furthermore, the edge preprocessing includes at least one of Kalman filtering, wavelet denoising, and median filtering; the synchronization includes at least one of interpolation resampling and time window alignment; and the multi-source fusion adopts a weighted multimodal fusion algorithm or a deep learning method combining a Long Short-Term Memory (LSTM) network and an attention mechanism model.

[0043] Furthermore, the warning and execution control includes triggering at least one of the following when the risk level reaches or exceeds a preset threshold: an audible and visual alarm, an electronic display screen information prompt, and a voice broadcast. It also involves linking at least one of the following actuators: a ventilation system, a lighting system, a traffic guidance device, and a blocking gate. The delay from the trigger command to the execution action does not exceed 100ms.

[0044] Compared with the prior art, the technical solution of this application has the following beneficial technical effects:

[0045] 1. The multi-source sensor acquisition module of this invention covers multi-modal sensors such as strain, displacement, cross-sectional convergence, seepage pressure, temperature and humidity, acoustic emission, and video images, which can comprehensively reflect the tunnel structure status and environmental conditions. By configuring local buffers and time compensation units at each sensing node, and adopting the IEEE 1588 precision time protocol, GPS timing, or hardware pulse synchronization, millisecond-level or sub-millisecond-level sampling time alignment is achieved, eliminating timing deviations between multiple types of sensors from the source. The edge computing processing module utilizes a heterogeneous architecture of ARM + FPGA / embedded GPU to run preprocessing algorithms such as Kalman filtering, wavelet denoising, and outlier removal, effectively suppressing noise and removing outliers. At the same time, data synchronization is ensured through interpolation, resampling, and time window alignment. Based on weighted multi-modal fusion or LSTM combined with attention mechanism, it generates working condition feature vectors, realizing high-precision, multi-dimensional, and low-latency tunnel operation data fusion, providing reliable and real-time consistent data support for subsequent risk assessment and control.

[0046] 2. The state perception and adaptive control module of this invention uses a risk level calculation formula to weight and fuse key indicators such as strain standard deviation, displacement standard deviation, acoustic emission event rate, and image change rate. It trains the module using historical working condition data to obtain weighting coefficients, ensuring the adaptability of model parameters and prediction accuracy. When the risk level reaches or exceeds level 3, the system automatically increases the sampling frequency to twice the original value and lowers the alarm trigger threshold, achieving dynamic optimization of monitoring accuracy and response speed. This mechanism can automatically allocate computing and communication resources for different risk stages during tunnel operation, improving monitoring sensitivity and early warning timeliness in cases of sudden working conditions and gradual damage, reducing missed alarms and delayed warnings caused by insufficient sampling or high thresholds, and enhancing tunnel safety assurance capabilities.

[0047] 3. The local early warning and execution module of this invention can directly trigger audible and visual alarms, electronic displays, and voice broadcasts on-site when the risk level meets the requirements. It also controls actuators such as ventilation, lighting, traffic guidance, and barrier gates via CAN or RS485 bus, completing actions within 100ms to ensure response time meets emergency needs. The anomaly self-diagnosis module detects sensor drift, system crashes, data loss, link anomalies, and processing unit overload in real time, calculating the health index HI. When HI falls below 0.7, a maintenance alarm is triggered. In case of partial sensor failure, redundant sensors or model-predicted data are used for compensation to ensure data continuity and system stability. The communication and remote interaction module integrates multiple links including industrial Ethernet, 5G, LoRa, and Wi-Fi Mesh, supporting automatic switching and providing breakpoint resume and CRC / MD5 verification capabilities to ensure data integrity and seamless synchronization with the remote platform. This design maintains local closed-loop monitoring and response capabilities even in the event of external communication interruptions or partial equipment failure, significantly improving the reliability and disaster resilience of the tunnel monitoring system. Attached Figure Description

[0048] Figure 1 This is an overall schematic diagram of a tunnel multi-source monitoring system with edge intelligence and autonomous control capabilities provided in Embodiment 1 of the present invention;

[0049] Figure 2 This is a sequence diagram of rapid event handling in Embodiment 1 of the present invention;

[0050] Figure 3 This is a schematic diagram of the internal structure and deployment strategy of the multi-source sensor acquisition module in Embodiment 1 of the present invention;

[0051] Figure 4 This is a hardware and software functional block diagram of the edge computing processing module in Embodiment 1 of the present invention;

[0052] Figure 5 This is a flowchart of the state perception and adaptive control process in Embodiment 1 of the present invention;

[0053] Figure 6 This is a functional diagram of the local early warning and execution module in Embodiment 1 of the present invention;

[0054] Figure 7 This is a logic block diagram of the anomaly self-diagnosis module in Embodiment 1 of the present invention;

[0055] Figure 8 This is a schematic diagram of the multi-link switching timing of the communication and remote interaction module in Embodiment 1 of the present invention;

[0056] Figure 9 This is a timing diagram of a tunnel multi-source monitoring method with edge intelligence and autonomous control capabilities according to Embodiment 2 of the present invention.

[0057] Figure 10 This is a schematic diagram illustrating the key steps of a tunnel multi-source monitoring method with edge intelligence and autonomous control capabilities according to Embodiment 2 of the present invention.

[0058] Figure 11 This is a logic diagram of the anomaly self-diagnosis and data compensation in Embodiment 2 of the present invention;

[0059] Figure 12 This is a timing diagram of communication interruption and recovery in Embodiment 2 of the present invention. Detailed Implementation

[0060] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0062] Example 1:

[0063] refer to Figures 1-8 A tunnel multi-source monitoring system with edge intelligence and autonomous control capabilities is disclosed, comprising a multi-source sensor acquisition module, an edge computing processing module, a state perception and adaptive control module, a local early warning and execution module, an anomaly self-diagnosis module, and a communication and remote interaction module. The modules communicate with each other via wired or wireless data links, collaboratively forming an integrated tunnel monitoring and control system for high-frequency dynamic operating conditions. Its specific components are as follows:

[0064] 1) Multi-source sensor acquisition module, used to synchronously sense tunnel structure and environmental parameters in multiple modes. The parameters include at least structural strain, structural displacement, cross-sectional convergence, seepage pressure, temperature and humidity, acoustic emission event characteristics, and video image data.

[0065] The multi-source sensor acquisition module includes multiple distributed sensor nodes. Each sensor node has a built-in high-precision clock unit, data cache unit, and time synchronization interface for millisecond-level or sub-millisecond-level time alignment with the edge computing processing module.

[0066] The time synchronization interface can use IEEE 1588 Precision Time Protocol (PTP), Global Positioning System (GPS) time synchronization or wired hardware pulse synchronization signal;

[0067] While collecting raw monitoring data, the sensor node writes a high-precision time tag into the data frame header to ensure the temporal consistency of data from different types of sensors in subsequent fusion analysis.

[0068] 2) Edge computing processing module, used to preprocess, reduce noise, extract features, synchronize sampling, and perform multi-source fusion analysis on the raw data transmitted by the multi-source sensor acquisition module at the tunnel site;

[0069] The edge computing processing module may include a heterogeneous computing platform based on ARM processors and field-programmable gate arrays (FPGAs), or an embedded graphics processing unit (GPU) with parallel computing capabilities; preprocessing includes, but is not limited to, Kalman filtering, wavelet denoising, and outlier removal; sampling synchronization includes converting data from different sampling frequencies to a unified fusion frequency through interpolation, resampling, and unified time window alignment; fusion analysis may be based on a weighted multimodal fusion algorithm, a long short-term memory network (LSTM), and a deep learning model with attention mechanisms to generate feature vectors representing the current operating conditions, providing input for state perception and control.

[0070] 3) Status perception and adaptive control module, used to identify the operating status of the tunnel structure in real time and dynamically adjust the operating parameters of the monitoring system based on the identification results;

[0071] Operating parameters include sensor sampling frequency, power consumption mode, alarm threshold, and data reporting frequency;

[0072] The state perception and adaptive control module can run a working condition risk assessment model based on multi-index fusion. This model obtains the risk level by calculating the weighted sum of indicators such as strain standard deviation, displacement standard deviation, acoustic emission event rate, and image change rate. The calculation formula is as follows: ;in, For strain standard deviation, For the standard deviation of displacement, The number of acoustic emission events per unit time. For the rate of change of the image, These are the weight coefficients obtained during model training;

[0073] When the risk level reaches the preset threshold, the module automatically increases the sampling frequency to twice the original value and lowers the alarm trigger threshold.

[0074] 4) Local early warning and execution module, used to directly trigger early warning and execute emergency control actions at the tunnel site when the edge computing processing module identifies abnormal working conditions or the risk level output by the state perception and adaptive control module reaches the alarm conditions;

[0075] The warning system includes audible and visual alarms, text information display, and broadcast prompts. The actions to be performed include starting and stopping the ventilation system, adjusting the lighting intensity, switching the speed limit sign, and controlling the blocking gate.

[0076] The local early warning and execution module is connected to the actuator via a low-latency industrial bus (including but not limited to CAN bus and RS485 bus) to ensure that the time delay of command transmission and response does not exceed 100ms.

[0077] 5) Anomaly self-diagnosis module, used to detect the operating status of each sensor node and data link of the system in real time, and take isolation, switching or compensation measures when anomalies are detected; the detection includes sensor drift detection (based on Z-score judgment), timeout without data packet detection and data packet sequence number loss detection;

[0078] When a sensor fails or the data link is interrupted, the module automatically calls redundant sensor data or performs interpolation compensation based on historical data and predicted values ​​from the operating condition model, and updates the Health Index (HI). When the Health Index is below 0.7, the module outputs a maintenance prompt and stores the information in the local log and uploads it to the remote platform.

[0079] 6) Communication and remote interaction module, used to establish a data interaction channel between the tunnel site and the remote monitoring platform; the interaction channel may include wired communication links (industrial Ethernet) and wireless communication links (5G cellular network, LoRa self-organizing network or Wi-Fi Mesh).

[0080] The communication and remote interaction module has a multi-link automatic switching function. When the main link is interrupted, it automatically switches to the backup link and performs breakpoint resume transmission after the link is restored. To ensure data integrity, the module stores cached data in segments and timestamps it before uploading, and uses Cyclic Redundancy Check (CRC) or Message Digest (MD5) for verification. It supports local storage and seamless synchronous uploading of more than 1 hour of continuous data.

[0081] As a preferred embodiment, a multi-source sensor acquisition module is used to synchronously acquire structural state parameters and environmental state parameters in multiple modes within the tunnel monitoring area. The parameters include at least structural strain, structural displacement, cross-sectional convergence, seepage pressure, temperature and humidity, acoustic emission event characteristics, and video image data.

[0082] The multi-source sensor acquisition module includes several distributed sensor nodes, a time synchronization unit, a data caching and management unit, and an anti-interference unit. Each sensor node is connected to the edge computing processing module via wired or wireless communication, forming a front-end acquisition network for multi-source heterogeneous data.

[0083] 1) Regarding the type and function of the sensing nodes, the sensing nodes include at least one or more of the following:

[0084] Strain sensing nodes: Fiber optic grating (FBG) sensors or resistance strain gauges are used to monitor minute strain changes in tunnel lining or surrounding rock.

[0085] Displacement sensing nodes: Laser displacement gauges, wire displacement gauges, or optical ranging modules are used to monitor the relative displacement of the tunnel structure;

[0086] Convergence sensing node: The convergence deformation of the cross section is obtained based on multi-point ranging or three-dimensional laser scanning technology;

[0087] Pressure sensing nodes: Vibrating wire or MEMS pressure sensors are used to monitor changes in water pressure behind the surrounding rock or lining;

[0088] Temperature and humidity sensing node: Based on digital temperature and humidity sensors, measure microclimate parameters inside the tunnel;

[0089] Acoustic emission sensing node: Piezoelectric or fiber optic acoustic emission probes are used to monitor microcrack propagation or rock mass fracturing events;

[0090] Video image sensing node: Low-light high-definition camera or infrared thermal imager is used to acquire visible light images or temperature distribution on the tunnel surface.

[0091] 2) Regarding the node deployment strategy, to ensure monitoring coverage and data representativeness, the sensor nodes are deployed according to the following principles:

[0092] Strain and displacement sensing nodes are installed at key structural parts of the tunnel (including the arch crown, arch waist, arch foot, tunnel floor, and areas of concentrated deformation).

[0093] Install seepage pressure sensing nodes in areas with high permeability surrounding rock, at joints, and at potential seepage points;

[0094] Deploy temperature and humidity sensing nodes in areas with significant temperature and humidity changes or those prone to condensation or freezing damage.

[0095] Acoustic emission sensor nodes are deployed in locations where the structure is prone to sudden damage (such as fault zones and weak interlayer areas);

[0096] Video image sensor nodes are deployed along major traffic flows, in key operational locations, and in areas prone to lining spalling or leakage.

[0097] 3) Time synchronization and calibration: The multi-source sensor acquisition module has a built-in time synchronization unit to achieve a unified time reference across sensor nodes. The time synchronization unit includes:

[0098] Time source: Can be a Global Positioning System (GPS), BeiDou Navigation Satellite System (BDS) or IEEE 1588 Precision Time Protocol (PTP) clock;

[0099] Synchronization method: Achieving a time synchronization error of no more than 1ms through hardware pulse synchronization signal (1PPS), network time synchronization (NTP / PTP) or fiber optic distributed clock signal;

[0100] Calibration method: Perform multiple rounds of synchronous calibration during system initialization, combined with the sensor response delay compensation formula.

[0101] ;

[0102] in, To calibrate the timestamp, This is the original data collection time. For sensor response delay, This refers to communication transmission delay.

[0103] 4) Data caching and management: The data caching and management unit includes a circular buffer and a data queue manager, which are used to store at least 1 minute of raw data to prevent data loss due to short-term communication interruptions. The data buffer adds sensor ID, timestamp, data type and data quality identifier (DQI) to each data packet to facilitate rapid indexing and fusion by the edge computing processing module.

[0104] 5) Anti-interference design: To adapt to the complex electromagnetic environment inside the tunnel, the multi-source sensor acquisition module adopts the following anti-interference measures: wired communication uses shielded twisted pair or optical fiber transmission, combined with grounding isolation and surge protection; wireless communication uses frequency hopping switch (FHSS) and channel coding (such as LDPC) to improve anti-interference capability; the sensor node shell adopts a waterproof, dustproof and corrosion-resistant design, with a protection level of not less than IP65.

[0105] 6) Functional scalability: The multi-source sensor acquisition module supports hot-swappable sensor replacement and online node expansion. New nodes can be quickly added to the existing monitoring network through adaptive addressing and automatic synchronization algorithms, and time synchronization and data access can be completed within 30 seconds.

[0106] As a preferred embodiment, the edge computing processing module is used to process the raw monitoring data transmitted by the multi-source sensor acquisition module in real time at the tunnel site, so as to realize data preprocessing, sampling synchronization, feature extraction and multi-source fusion analysis, and provide the analysis results to the state perception and adaptive control module and the local early warning and execution module. The module includes hardware units, software processing units, data storage units and security units. The units work together to achieve low latency and high reliability edge-side intelligent computing.

[0107] 1) Hardware unit, processing core: The hardware unit includes a heterogeneous computing architecture consisting of at least one multi-core ARM processor and a field-programmable gate array (FPGA) or embedded graphics processing unit (GPU). The ARM processor is used to perform general data processing tasks, and the FPGA or GPU is used to accelerate filtering, feature extraction and deep learning inference operations in parallel.

[0108] Interface modules include a multi-channel high-speed data acquisition card (ADC / DAC), a gigabit Ethernet interface, a wireless communication interface (5G, Wi-Fi, LoRa), a USB interface, and a serial bus interface (RS485, CAN).

[0109] Local storage: Uses solid-state storage (SSD) or eMMC chips with a capacity of no less than 64GB to cache raw data and processing results, and supports breakpoint resume and historical data rollback;

[0110] Clock unit: Built-in high-precision temperature-compensated crystal oscillator (TCXO) and optional GPS / BDS timing module to maintain the time accuracy of edge processing to no more than ±1ms.

[0111] 2) Software processing unit, including preprocessing module, synchronization module, feature extraction module, and multi-source fusion module:

[0112] Preprocessing module: used to suppress noise and remove outliers from the raw data. Noise suppression includes Kalman filtering, wavelet denoising and median filtering. Outlier removal is based on the statistical 3σ principle or the sliding window Z-score algorithm.

[0113] Synchronization module: Based on the timestamps and sampling frequencies of each sensor, it performs interpolation resampling and time window alignment on different types of data, mapping heterogeneous data onto a unified fusion time axis;

[0114] Feature extraction module: Differentiated algorithms are used to extract features for different signal types. For example, statistical features such as mean, variance, and peak frequency are extracted from strain and displacement signals; features such as event count, amplitude distribution, and dominant frequency are extracted from acoustic emission signals; and deep feature vectors based on convolutional neural networks (CNN) are extracted from video images.

[0115] Multi-source fusion module: Employs a weighted multimodal fusion algorithm or a deep learning model combining Long Short-Term Memory (LSTM) network and attention mechanism to perform temporal correlation analysis on features of different modalities and output a comprehensive feature vector representing the tunnel working conditions.

[0116] 3) Data processing workflow and performance indicators. The data processing workflow includes:

[0117] Step 1: Receive the raw data packet with high-precision time stamp transmitted by the multi-source sensor acquisition module;

[0118] Step 2: Perform preprocessing and synchronization operations on the FPGA / GPU, and send the results to the ARM processor;

[0119] Step 3: Perform feature extraction and multi-source fusion on the ARM processor;

[0120] Step 4: Write the fusion results into local storage in the form of structured data, and send them to the state awareness and adaptive control module through a shared memory mechanism;

[0121] The processing performance indicators include: ① Single data packet processing latency not exceeding 50ms; ② Support for parallel access and processing of no less than 100 channels of sensor data; ③ Data packet loss rate not exceeding 0.01%; ④ CPU and acceleration unit utilization not less than 70%.

[0122] 4) Regarding redundancy and reliability design, including:

[0123] Hardware redundancy: The processing core supports dual-machine hot backup mode. When the main processing unit fails, the backup processing unit takes over the computing task within 5 seconds.

[0124] Data redundancy: Critical processing results are written in duplicate, and data integrity is verified by CRC or MD5 checksum.

[0125] Fault tolerance mechanism: When an anomaly occurs in an individual data channel (such as signal loss or time drift), the system automatically calls the prediction model (such as the ARIMA model based on historical data) to generate compensation data in order to maintain the continuity of fusion operation.

[0126] 5) Security Unit

[0127] Data encryption: All transmitted data packets are encrypted using AES-256 symmetric encryption to prevent the communication link from being illegally intercepted;

[0128] Access control: Restricting unauthorized users from accessing processing units through role-based access control (RBAC) mechanisms;

[0129] Log auditing: Automatically generates processing logs, including timestamps, input data sources, processing algorithm version numbers, and output result summaries, facilitating post-event traceability and review.

[0130] Preferably, the state perception and adaptive control module is used to identify the tunnel structure's operating status and environmental status in real time, and dynamically adjust the operating parameters of the multi-source sensor acquisition module, edge computing processing module, and local early warning and execution module based on the identification results to adapt to different levels of dynamic working condition changes; the module includes a state identification unit, a risk assessment unit, an adaptive control unit, and a strategy management unit.

[0131] 1) The status identification unit is used to receive the multi-source fusion feature vector output by the edge computing processing module, and combine it with historical data, benchmark models and current monitoring values ​​to determine the current operating status of the tunnel;

[0132] The state recognition can be based on a state machine model, which includes at least four states: normal operation, mild abnormality, moderate abnormality, and severe abnormality.

[0133] The determination process includes feature normalization, comparison with state boundary thresholds, and state transition logic judgment. The state transition rules can be based on: ;

[0134] in, For the state at the next moment, This is the current state. For the current feature vector, This is the threshold parameter set.

[0135] 2) The risk assessment unit is used to calculate the risk level value of the current operating condition. This allows for the quantification of potential security risks under different conditions; risk level calculation can employ a multi-indicator weighted fusion formula: ;

[0136] in, For strain standard deviation, For the standard deviation of displacement, The rate of acoustic emission events per unit time. The rate of change between adjacent image frames. These are the weight coefficients obtained through training with historical operating data;

[0137] The risk level R is mapped to five risk levels (level 1 is the lowest and level 5 is the highest) and output to the adaptive control unit.

[0138] 3) Adaptive control unit, which dynamically adjusts system operating parameters based on the risk level value and status identification results output by the risk assessment unit, including:

[0139] Sampling frequency: At low risk levels (levels 1-2), the sampling frequency can be reduced to 50%-80% of the baseline value to save energy; at high risk levels (levels 4-5), the sampling frequency can be increased to 150%-200% of the baseline value to enhance monitoring resolution.

[0140] Power consumption mode: Under low-risk conditions, some sensor nodes enter low-power sleep mode, while under high-risk conditions, all nodes are activated and running.

[0141] Alarm threshold: The alarm trigger value of each monitored quantity is dynamically adjusted according to the risk level, making the alarm more sensitive under high-risk conditions;

[0142] Data upload frequency: When the risk is low, extend the batch data upload cycle (e.g., once every 10 minutes); when the risk is high, shorten it to real-time or second-level upload.

[0143] Control commands are sent to the corresponding modules via wired or wireless links, requiring parameter switching to be completed within a specified delay time (not exceeding 200ms).

[0144] 4) The strategy management unit is used to manage the rule set and strategy version of adaptive regulation, and supports online strategy updates and historical strategy rollback. The strategy management unit includes a strategy library, a version control module and a rollback module. When a new regulation strategy fails to load at the edge or causes an exception, it can be rolled back to the previous version within 1 second. The strategy switching process adopts a smooth transition method (such as linear interpolation or gradual increase / decrease control) to avoid system oscillation caused by sudden changes in sampling frequency or threshold.

[0145] 5) Redundancy and security mechanisms

[0146] Redundant computation: The risk assessment results are calculated in parallel by the main computation unit and the redundant computation unit, and the reliability of the computation is verified by the consistency test of the results (e.g., the difference does not exceed 5%).

[0147] Security Lockout: When the system detects malicious interference signals or signs of data tampering, it immediately enters security lockout mode to maintain critical parameters within a safe range and disable remote parameter write permissions.

[0148] Emergency response to communication interruption: In the event of a communication interruption, the module automatically saves the most recent control strategy and continues to execute it until communication is restored.

[0149] As a preferred embodiment, the local early warning and execution module is used to realize rapid alarm and emergency response execution for abnormal working conditions at the tunnel site. When the edge computing processing module identifies abnormal working conditions or the risk level output by the state perception and adaptive control module reaches or exceeds the preset threshold, this module can operate independently without relying on remote cloud commands, achieving millisecond-level response. The module includes an early warning unit, an execution control unit, an interface conversion unit, and a safety protection unit.

[0150] 1) The early warning unit is used to send multi-mode alarm signals to tunnel site workers, operation and management personnel, and passing vehicles;

[0151] The warning methods include: ① Audible and visual alarm: using high-brightness LED flashlights and high-decibel buzzers, with light intensity ≥2000cd and sound pressure level ≥90dB, supporting dynamic adjustment of flashing frequency and sounding mode; ② Text and graphic information display: displaying the warning level, location, time, and handling suggestions in real time through electronic displays or LED matrix screens; ③ Voice broadcast: playing voice warnings through a tunnel broadcasting system, supporting Mandarin and multilingual switching;

[0152] The early warning unit supports a tiered triggering mechanism, with different risk levels corresponding to different warning intensities and information content. For example, a level 3 risk level only triggers an audible and visual alarm, while a level 5 risk level triggers an audible and visual alarm, a display, and a voice broadcast simultaneously.

[0153] 2) The execution control unit is used to directly link the tunnel's electromechanical and safety facilities in high-risk or emergency situations to execute emergency response actions;

[0154] The actions to be performed include, but are not limited to: ① Ventilation system control: starting or stopping jet fans, adjusting ventilation direction and air volume; ② Lighting system adjustment: increasing or decreasing lighting brightness, switching emergency lighting modes; ③ Traffic guidance and closure: controlling variable message signs, speed limit signs, traffic lights or automatic gates to guide traffic flow and restrict passage; ④ Blocking and drainage: activating fireproof water curtains, closing fire doors, starting or stopping drainage pumps;

[0155] The execution control unit communicates with the actuator through a low-latency industrial bus, including but not limited to CAN bus, RS485 bus and Ethernet control interface, with a total delay of no more than 100ms for a single execution command transmission and action response.

[0156] 3) The interface conversion unit is used to achieve communication compatibility between the local early warning and execution module and various execution devices. It supports digital I / O, analog signals, level conversion and multi-protocol conversion. When the execution device is working under different communication protocols (such as Modbus, CANopen, Profibus), the interface conversion unit can automatically identify the protocol type and load the corresponding driver to achieve plug and play.

[0157] 4) The safety protection unit is used to prevent false alarms or erroneous command execution from affecting tunnel operations, and includes:

[0158] Dual triggering mechanism: The execution of an action must simultaneously satisfy both the risk level judgment signal and the equipment health status confirmation signal to avoid triggering erroneous actions due to sensor failure;

[0159] Manual intervention interface: Allows authorized operators to manually confirm, delay, or suspend alerts and actions via a local console or mobile terminal;

[0160] Emergency manual mode: In the event of a failure of the automatic control system or a complete power outage, key actions (such as emergency lighting, ventilation, and drainage) can be directly triggered by a manual button driven by a mechanical or independent power source.

[0161] Self-sustaining in case of communication link interruption: Continue to execute the last effective control strategy until communication is restored or manual takeover is required.

[0162] 5) Redundancy and reliability design, hardware redundancy: the sound and light alarm and the broadcast system adopt dual-circuit power supply, and the execution control unit adopts dual-channel output. When the main channel fails, the backup channel takes over within 50ms.

[0163] Power supply guarantee: Equipped with an uninterruptible power supply (UPS) or an independent battery pack, this module can maintain continuous operation for no less than 2 hours after an external power outage;

[0164] Self-testing mechanism: The module automatically performs a full-function self-test every 24 hours, including audible and visual alarm tests, actuator continuity tests, and interface protocol handshake tests, and stores the test results in local logs and uploads them to a remote platform.

[0165] Preferably, the anomaly self-diagnosis module is used to perform real-time health monitoring and fault diagnosis on the operating status of the multi-source sensor acquisition module, edge computing processing module, status perception and adaptive control module, local early warning and execution module, and communication link during the operation of the tunnel monitoring system. Upon detecting anomalies, it automatically performs isolation, switching, compensation, and recovery operations to ensure that the system maintains continuous monitoring and emergency response capabilities under dynamic operating conditions and partial functional failures. This module includes a health monitoring unit, an anomaly detection unit, a fault isolation and switching unit, a data compensation unit, and a diagnostic recording and reporting unit.

[0166] 1) The health monitoring unit is used to collect and analyze the operating status parameters of each module. The parameters include at least the sensor output frequency, signal amplitude stability, data packet integrity, communication link delay, CPU / GPU utilization, memory usage and device temperature.

[0167] The monitoring frequency can be adaptively adjusted within the range of 1 Hz to 0.01 Hz according to the system operating conditions;

[0168] The unit compares the data with device feature templates using a sliding window statistical method to generate a real-time health index (HI), calculated as follows: ;

[0169] in, The current parameter value. For reference only. For the acceptable range of parameter variation, These are the weighting coefficients.

[0170] 2) The anomaly detection unit is used to determine whether there are anomalies in each module based on real-time health indices and a specific rule base. Detection types include:

[0171] Sensor drift detection: The Z-score method or multi-point comparison method is used to identify situations where the output deviates from the benchmark for a long period of time;

[0172] Sensor crash detection: Monitors the data packet reception interval. If the number of consecutively lost data packets exceeds a preset threshold (e.g., 3 sampling periods), it is determined to be a crash.

[0173] Data loss detection: Detects missing or corrupted data using data packet sequence numbers and CRC checksums;

[0174] Communication link anomaly detection: Monitor round-trip time (RTT), packet loss rate and link jitter, and determine instability or interruption if the threshold is exceeded;

[0175] Processing unit overload detection: Monitors the CPU / GPU utilization of the edge computing processing module, and triggers an overload alarm when it continuously exceeds 90% and the latency exceeds the limit.

[0176] 3) The fault isolation and switching unit is used to isolate the abnormal device or channel from the system and activate the redundant device or backup channel when an anomaly is detected.

[0177] Isolation strategies include hardware isolation (cutting off power or data bus), logical isolation (shielding abnormal node data), and protocol isolation (blocking abnormal node communication).

[0178] Redundancy switching includes: ① Sensor redundancy: Two sensors of different types or different models of the same type are deployed at the same monitoring point. When the main sensor fails, the switch is made to the backup sensor; ② Communication redundancy: When the main link (such as fiber optic) fails, the switch is made to the backup link (such as 5G or LoRa) with a switching time of no more than 5 seconds.

[0179] 4) The data compensation unit is used to generate compensation data through algorithms in the event of partial data loss or delay, ensuring the normal operation of subsequent analysis and control modules;

[0180] The compensation strategies include: ① Interpolation compensation: linear interpolation, spline interpolation or polynomial fitting are used to generate compensation values ​​for short-term missing data (duration <10 seconds); ② Model prediction compensation: for long-term missing data, prediction data are generated based on time series prediction models (such as ARIMA, LSTM) trained on historical data; ③ Neighbor node inference compensation: spatial correlation is used to infer the possible values ​​of missing nodes through data from adjacent sensor deployments.

[0181] 5) The diagnostic recording and reporting unit is used to record abnormal events, diagnostic results, compensation data and handling measures to local storage, and upload them to the remote monitoring platform in batches when the communication link is available. The recorded content includes the event occurrence time, module name, abnormality type, handling measures, compensation algorithm type and health index change curve. This unit supports two modes: instant reporting based on event trigger and scheduled batch reporting. It can also send key abnormal information directly to maintenance personnel via SMS, email or dedicated push channel.

[0182] 6) Regarding safety and reliability assurance, the anomaly self-diagnosis module has a dual-processing channel redundancy mechanism. The main channel is responsible for real-time diagnosis, and the secondary channel performs low-frequency health checks. When the difference between the two diagnostic results exceeds the preset range, a manual review process is triggered. When communication is interrupted, the anomaly self-diagnosis module continues to perform local diagnosis and data compensation until communication is restored. The anomaly self-diagnosis module supports remote updates of the strategy library to ensure that the anomaly detection algorithm can be iteratively optimized according to new fault modes.

[0183] Preferably, the communication and remote interaction module is used to establish a stable, low-latency, and highly reliable two-way data transmission channel between the tunnel monitoring site and the remote monitoring center, enabling real-time reporting of monitoring data, issuance of control commands, and remote maintenance of system operation status. This module includes a multi-link communication unit, a protocol and data management unit, a link failure detection and recovery unit, a security and encryption unit, and a communication redundancy unit.

[0184] The multi-link communication unit has both wired and wireless communication capabilities, supporting various link types such as industrial Ethernet, fiber optic communication, 5G cellular communication, LoRa self-organizing network, and Wi-Fi Mesh. In wired communication, a gigabit Ethernet interface or fiber optic transceiver module can be used to achieve a transmission rate of no less than 1 Gbps. In wireless communication, public 5G, private LTE, LoRa, or Wi-Fi Mesh can be selected according to the actual tunnel conditions to meet different transmission distance and bandwidth requirements. The communication and remote interaction module has a built-in link status monitoring circuit, which detects parameters such as link delay (RTT), packet loss rate, and signal strength (RSSI) in real time, and outputs the link quality index (LQI).

[0185] The protocol and data management unit is used to achieve protocol compatibility and data transmission optimization across different links and upper-level platforms, supporting MQTT, Modbus TCP, IEC 104, HTTPS, and dedicated encrypted transmission protocols. Data management adopts a hierarchical caching mechanism, including a real-time cache and a historical cache. The real-time cache is used for critical data reported at the millisecond level, while the historical cache is used for non-critical data uploaded in batches. When data is segmented and packaged, timestamps, data type identifiers, data quality identifiers (DQI), and cyclic redundancy check codes (CRC) are attached to ensure data integrity and traceability.

[0186] Both data reporting and transmission use an acknowledgment / reply mechanism (ACK / NACK). If no acknowledgment signal is received, the data will be automatically retransmitted. The number of retransmissions is configurable (default 3 times).

[0187] The link failure detection and recovery unit detects the communication link status through periodic heartbeat packets. When no heartbeat response is received for n consecutive times (n≥3), it is determined that the link is interrupted.

[0188] The chain break recovery unit has the following functions:

[0189] Link switching: Automatically switches to a backup link (such as fiber optic → 5G, or 5G → LoRa) after the primary link is interrupted, with a switching time of no more than 5 seconds;

[0190] Resumable interruption: During the link failure, cached data is saved to local storage in chronological order and uploaded in batches in timestamp order after the link is restored, ensuring data integrity and correct timing;

[0191] Data priority scheduling: After recovery, high-priority data (such as changes in risk level, alarm events) will be uploaded first, and the remaining data will be transmitted in time batch order.

[0192] Security and encryption units are used to prevent data leakage and tampering during communication, including:

[0193] Data encryption: AES-256 symmetric encryption or RSA asymmetric encryption is used to ensure the confidentiality of transmitted data;

[0194] Identity authentication: Verifying the identities of both communicating parties through a two-way authentication mechanism based on digital certificates (such as TLS / SSL);

[0195] Data integrity verification: The SHA-256 digest algorithm is used to sign and verify the data packets to ensure that the data has not been modified during transmission;

[0196] Anti-replay mechanism: embeds random numbers and timestamps into data packets to prevent old data from being maliciously resent.

[0197] The communication redundancy unit is used to improve the system's continuous operation capability in the event of a link failure, including: ① Multi-link load balancing: When multiple links are available simultaneously, data traffic is dynamically allocated based on bandwidth and latency; ② Redundant link hot backup: The backup link maintains a low-frequency heartbeat connection with the remote platform in real time to ensure immediate takeover after the main link is interrupted; ③ Hierarchical redundancy strategy: Different redundancy strategies are set for different types of data. For example, alarm data is sent simultaneously through two links, while normal status data is sent only through the main link.

[0198] Regarding remote interaction functionality, in addition to data transmission, the communication and remote interaction module also supports remote maintenance and interaction, including: ① Remote parameter distribution: receiving parameter update commands such as sampling frequency, alarm threshold, and power consumption mode from the remote monitoring platform; ② Remote diagnostics: uploading the diagnostic results of the anomaly self-diagnosis module in real time, supporting remote distribution of test and reset commands; ③ Firmware and policy updates: receiving system firmware and control policy update files through a secure transmission channel, and automatically loading them after verification; ④ Two-way message confirmation: remotely distributed commands must be confirmed locally and an execution receipt must be returned to ensure the reliability and traceability of command issuance.

[0199] Example 2:

[0200] refer to Figures 9-12 A tunnel multi-source monitoring method with edge intelligence and autonomous control capabilities is applied to the above-mentioned system, comprising the following steps:

[0201] S1. Multi-source data acquisition and high-precision synchronization: At the tunnel structure monitoring site, multi-source sensor acquisition modules are used to acquire multi-modal monitoring data such as structural strain, structural displacement, cross-sectional convergence, seepage pressure, temperature and humidity, acoustic emission signals, and video images in real time. Each sensor node in the multi-source sensor acquisition module has a built-in high-precision clock unit, and a unified time reference is achieved through Precision Time Protocol (PTP), Global Positioning System (GPS) timing signals, or hardware pulse (1PPS) signals. High-precision time tags are added to the acquired raw data, and the time synchronization error is controlled within ±1 millisecond to ensure the consistency of time sequence of different types of data in subsequent fusion analysis.

[0202] S2. Edge Data Preprocessing and Fusion: Input the time-stamped raw data uploaded by the multi-source sensor acquisition module into the edge computing processing module; perform data preprocessing operations at the edge, including Kalman filtering, wavelet denoising, and outlier removal based on sliding window, to eliminate sensor noise and isolated erroneous data; perform interpolation resampling and unified time window alignment on data with different sampling frequencies to map heterogeneous data to a unified fusion time axis;

[0203] Differentiated features are extracted for different modal data, such as statistical features of strain / displacement signals, event frequency and main frequency of acoustic emission signals, and deep convolutional feature vectors of video images; a comprehensive feature vector representing the current tunnel condition is generated based on a weighted multimodal fusion algorithm or a deep learning model combining a long short-term memory network (LSTM) with an attention mechanism.

[0204] S3. State Perception and Risk Level Determination; The comprehensive feature vector is input into the state perception and adaptive control module; based on the operating condition state machine model (including normal operation, minor anomaly, moderate anomaly, severe anomaly, etc.), combined with the risk assessment formula: ;in, For strain standard deviation, For the standard deviation of displacement, For acoustic emission event rate, The rate of change between adjacent frames. The preset weighting coefficients are used to calculate the risk level R, which is then mapped to a five-level risk level (level 1 is the lowest and level 5 is the highest).

[0205] S4. Adaptive Sampling and Threshold Adjustment: When the risk level is low (level 1-2), reduce the sensor sampling frequency to 50%-80% of the base value, some sensors enter low-power mode, and increase the data upload interval to reduce communication load; when the risk level is medium (level 3), maintain the base sampling frequency and appropriately tighten the alarm threshold; when the risk level is high (level 4-5), increase the sampling frequency to 150%-200% of the base value, all sensor nodes remain active, and lower the alarm threshold to enhance sensitivity, and shorten the data upload cycle to the second level or real-time mode; the adjustment command is sent to the multi-source sensor acquisition module and edge computing processing module through the edge control link, with a delay of no more than 200 milliseconds.

[0206] S5. Local rapid early warning and execution control: When the risk level reaches or exceeds the set early warning threshold, the local early warning and execution module immediately activates multi-mode alarms, including audible and visual alarms, electronic displays, and voice broadcasts; at the same time, according to the preset emergency linkage strategy, it directly controls the electromechanical equipment in the tunnel, including starting or stopping the jet fan, adjusting the lighting brightness, switching speed limit signs, opening the fireproof water curtain, or blocking the gate; the execution commands are issued through a low-latency industrial bus (such as CAN, RS485), and the delay of a single command transmission and execution does not exceed 100 milliseconds.

[0207] S6. Anomaly self-diagnosis and data compensation: The anomaly self-diagnosis module monitors the operating status of each sensor, edge processing unit, execution control unit and communication link in real time, and detects anomaly types including sensor drift, system crash, data loss, link interruption and processing unit overload.

[0208] When an anomaly is detected, fault isolation and redundancy switching strategies are automatically executed, such as switching to a backup sensor or backup communication link; for missing data, interpolation compensation (short-term missing), model prediction compensation (long-term missing), or neighbor node inference compensation (spatial correlation) methods are used to generate compensation data to ensure data continuity and the reliability of operating condition judgment; diagnostic results and compensation data are recorded in the local log and uploaded to the remote platform after communication is restored.

[0209] S7. Communication Recovery and Remote Synchronization: When the communication and remote interaction module detects a link interruption, it activates a backup link or caches data to local storage. After the link is restored, it performs breakpoint resumption and uploads the cached data in batches according to timestamp order to the remote monitoring platform. Before uploading, it performs integrity verification (CRC, MD5, or SHA-256) on the data to ensure that the data is not corrupted during transmission.

[0210] The remote monitoring platform archives, visualizes, and performs historical analysis on the received data, and can send parameter updates and adjust strategies based on the latest status.

[0211] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A tunnel multi-source monitoring system with edge intelligence and autonomous control capabilities, characterized in that, include: The multi-source sensor acquisition module is used to acquire multimodal parameters of the tunnel structure and environment, and attach high-precision time tags at the moment of acquisition; The edge computing processing module is used to perform real-time preprocessing, sampling synchronization, and multi-source fusion of the raw data from the multi-source sensor acquisition module at the tunnel site. The status awareness and adaptive control module is used to identify the tunnel operating status in real time and dynamically adjust the sampling frequency, power consumption mode, alarm threshold and data upload frequency based on the operating condition feature vector output by the edge computing processing module. The local early warning and execution module is used to directly trigger an early warning and link the execution mechanism at the tunnel site when the risk level output by the state perception and adaptive control module reaches a preset threshold. The anomaly self-diagnosis module is used to monitor the operating status of each module in the system in real time and to isolate, switch or compensate data when a fault occurs. The communication and remote interaction module is used to interact with the remote platform via wired and wireless multi-links, and to perform breakpoint resume and data integrity verification after communication is restored.

2. The system according to claim 1, characterized in that, The multi-source sensor acquisition module includes strain sensors, displacement sensors, cross-sectional convergence monitoring sensors, pressure sensors, temperature and humidity sensors, acoustic emission sensors, and video image sensors. Each sensing node has a built-in local cache and time compensation unit, and achieves millisecond-level or sub-millisecond-level time alignment through the IEEE 1588 precision time protocol, GPS timing, or hardware pulse synchronization signals.

3. The system according to claim 1, characterized in that, The edge computing processing module adopts a heterogeneous architecture of ARM processor and FPGA or embedded GPU, runs preprocessing algorithms including Kalman filtering, wavelet denoising and outlier removal, uses interpolation, resampling and time window alignment to achieve sampling synchronization, and generates working condition feature vectors based on weighted multimodal fusion or LSTM combined with attention mechanism model.

4. The system according to claim 1, characterized in that, The state perception and adaptive control module calculates the risk level using the following formula: ； in, For strain standard deviation, For the standard deviation of displacement, For acoustic emission event rate, For the rate of change of the image, The weighting coefficients are obtained from training with historical data; when the risk level is ≥3, the sampling frequency is increased to twice the original value and the alarm trigger threshold is lowered.

5. The system according to claim 1, characterized in that, The local early warning and execution module includes an audible and visual alarm, an electronic display screen, a voice broadcasting unit, and an execution control unit connected to a ventilation system, a lighting system, a traffic guidance device, and a barrier gate. The execution control unit communicates with the actuator via a CAN bus or an RS485 bus and completes the action within 100 ms after receiving a trigger signal.

6. The system according to claim 1, characterized in that, The abnormal self-diagnosis module determines faults by detecting sensor drift, system crash, data loss, communication link abnormalities, and processing unit overload, calculates the health index HI, triggers a maintenance alarm when HI < 0.7, and calls redundant sensor data or data based on model prediction for compensation when some sensors fail. The communication and remote interaction module includes industrial Ethernet, 5G communication, LoRa and Wi-Fi Mesh communication units. It has a multi-link automatic switching function, switches to the backup link when the main link is interrupted, and performs breakpoint resume transmission based on the timestamp index after the link is restored. It uses CRC or MD5 check to ensure data integrity and supports local storage and seamless synchronization of at least 1 hour of continuous data.

7. A tunnel multi-source monitoring method with edge intelligence and autonomous control capabilities, characterized in that, include: Collect multi-source sensor data within the tunnel monitoring area and synchronize the data in time. At the edge, preprocessing, synchronization, and multi-source fusion of the collected data are performed to generate characteristic data representing the working conditions. Calculate the risk level based on the aforementioned feature data; The sampling frequency, power consumption mode, alarm threshold, and data reporting frequency of the monitoring system are dynamically adjusted according to the risk level. When the risk level reaches the preset conditions, an early warning is triggered at the tunnel site and the execution mechanism is activated. The system performs anomaly detection during operation and isolates, switches, or compensates for data when a fault occurs. When the communication link is restored, perform breakpoint resume transmission, integrity verification and remote synchronization of data.

8. The method according to claim 7, characterized in that, The risk level calculation steps include: Based on strain standard deviation standard deviation of displacement Acoustic emission event rate and image change rate Calculate risk level value The calculation formula is as follows: ； in, These are the weight coefficients obtained by training based on historical operating data.

9. The method according to any one of claims 7 or 8, characterized in that, The edge preprocessing includes at least one of Kalman filtering, wavelet denoising, and median filtering; the synchronization includes at least one of interpolation resampling and time window alignment; the multi-source fusion adopts a weighted multimodal fusion algorithm or a deep learning method combining a Long Short-Term Memory (LSTM) network and an attention mechanism model.

10. The method according to any one of claims 7 or 8, characterized in that, The warning and execution control includes triggering at least one of the following when the risk level reaches or exceeds a preset threshold: audible and visual alarm, electronic display screen information prompt, and voice broadcast; and linking at least one of the following actuators: ventilation system, lighting system, traffic guidance device, and blocking gate; and the delay from triggering command to execution action shall not exceed 100 ms.

Images