Wireless transmission anchor rod multipoint stress sensor system

Abstract

The invention discloses a wireless transmission anchor rod multipoint stress sensor system, which relates to the technical field of intelligent monitoring of geotechnical engineering and comprises a data sensing and acquisition module, a data processing and transmission module, a data receiving and storage module, a data analysis and decision module and a user interaction and management module. The technical key points are as follows: multi-protocol adaptive communication and quantum encryption are carried out, a multi-protocol adaptive communication stack is based on an improved near-end strategy optimization algorithm, intelligent switching of communication protocols can be completed in an extremely short time according to an actual environment, and the intelligent switching of the communication protocols can be completed in scenes such as mountainous area tunnels with complex and changeable signals. The problem that transmission of a traditional single protocol is prone to being interrupted is effectively avoided, continuity of data transmission is guaranteed, meanwhile, the integrated quantum key distribution unit adopts a BB84 protocol and is combined with a two-factor encryption system, data transmission safety is remarkably improved, quantum attacks can be effectively resisted in a monitoring scene with the extremely high requirement for data safety in confidential work, and the security of data transmission is improved. And the data leakage risk is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of geotechnical engineering intelligent monitoring, and in particular to a wireless transmission anchor rod multi-point stress sensor system. Background Art

[0002] In modern geotechnical engineering construction and maintenance systems, anchor bolts serve as core support components for the stability of tunnels, slopes, foundation pits, and other structures. Accurate monitoring of their stress state is directly related to project safety and service life. However, currently widely used traditional monitoring methods have significant limitations in complex engineering scenarios.

[0003] At the data collection level, the drawbacks of traditional systems relying on a single communication protocol are particularly prominent. For example, in a highway tunnel project in the southwestern mountainous area, due to mountain obstruction and frequent geological activity, the data packet loss rate of the anchor stress monitoring system based on 4G communication remained at around 18% for a long time, severely compromising data continuity during the critical construction phase. In a slope monitoring scenario at a northern mine, the wired sensor network suffered line damage due to the freeze-thaw cycle of permafrost, resulting in annual maintenance costs reaching 12% of the total project budget. Furthermore, manual inspection cycles lasted up to 15 days, making it difficult to capture sudden stress changes. Furthermore, the traditional sensor calibration process is complex. For example, in a large-scale water conservancy project, manual calibration of a single anchor sensor took over four hours, with an overall error rate exceeding 1.2% FS, failing to meet the requirements of high-precision monitoring.

[0004] There are also significant defects in the data processing link. In the construction monitoring of coastal subway tunnels, the multi-sensor network lacks an effective spatiotemporal alignment mechanism, and the time synchronization error generally exceeds 15 microseconds, and the spatial positioning deviation reaches 25 centimeters. This makes it difficult to match and analyze stress data with geological radar images, delaying construction risk warnings. In a deep foundation pit project in a certain city, the traditional system is unable to integrate multi-source heterogeneous data such as anchor stress, groundwater level, and surrounding building settlement. When encountering continuous heavy rains, the support structure became partially unstable due to the failure to comprehensively analyze the relationship between the data, resulting in direct economic losses of approximately 8 million yuan.

[0005] Insufficient decision-making support capabilities are another shortcoming of traditional monitoring systems. A high-speed railway tunnel project used a fixed-threshold early warning mechanism. When crossing a fault fracture zone, it did not consider the dynamic changes in surrounding rock stress, resulting in three false alarms and two missed alarms, seriously disrupting the construction progress. In the maintenance scenario of old slopes, the decision-making model that relies on manual experience makes the allocation of maintenance resources inefficient. For example, a small-scale landslide occurred on the slopes along a provincial highway due to untimely maintenance, interrupting traffic for 36 hours and causing economic losses of more than 5 million yuan.

[0006] With the advancement of the "new infrastructure" strategy, the scale and complexity of geotechnical engineering projects continue to rise. Traditional monitoring technology can no longer meet the urgent needs of projects for high reliability, high intelligence, and high safety. The development of a new anchor stress monitoring system has become a key technological breakthrough direction to ensure the safety of major projects. Summary of the Invention

[0007] (1) Technical problems solved

[0008] In view of the deficiencies in the prior art, the present invention provides a wireless transmission anchor rod multi-point stress sensor system, which solves the problems raised in the background art.

[0009] (2) Technical solution

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0011] A wireless transmission anchor rod multi-point stress sensor system, the system comprising:

[0012] The data sensing and acquisition module establishes a two-way communication link with the anchor stress sensor based on an adaptive communication protocol stack, and uses a strategy combining polling and event triggering to collect data in real time. The protocol stack supports dynamic switching of multiple protocols.

[0013] The data processing and transmission module builds a three-level data cleaning pipeline. Based on the timestamp signal provided by the Beidou satellite timing system, it combines the seven-parameter Bursa-Wolf transformation model with the least squares matching algorithm to perform spatiotemporal alignment, extract multi-dimensional feature parameters, build a multi-protocol fusion transmission architecture, and perform lightweight data compression.

[0014] The data reception and storage module is deployed in the cloud or on a local server, equipped with a triple data verification mechanism, a distributed database architecture with a three-dimensional index structure based on the Hilbert space-filling curve, a hierarchical storage strategy, and remote disaster recovery with real-time log synchronization between the primary and standby databases based on the Paxos consensus algorithm.

[0015] The data analysis and decision-making module connects to the engineering knowledge base to build a hierarchical warning and maintenance decision-making model based on multi-objective optimization;

[0016] The user interaction and management module provides a multi-terminal adaptation interface, realizes 3D real-time rendering based on BIM model and WebGL technology, and integrates hierarchical permission management and system parameter management modules based on the RBAC model.

[0017] Furthermore, in the data sensing and acquisition module, the adaptive communication protocol stack adopts a protocol selection algorithm based on the Markov decision process, constructs a Q value matrix according to signal strength, packet loss rate, and available bandwidth, performs protocol switching through dynamic programming, and adopts a strategy combining polling and event triggering to collect data in real time, including stress values ​​and three-dimensional stress components;

[0018] The protocol stack supports multiple protocols including ZigBee and LoRa, and has a built-in sensor self-calibration program compliant with the IEEE1451 standard and a fault diagnosis module based on the local outlier factor algorithm and the autoregressive moving average model;

[0019] The sensor self-calibration program injects a step wave excitation signal, collects at least five sets of signal data with different excitation intensities, and uses the least squares curve fitting method to determine the calibration parameters by minimizing the sum of squared errors;

[0020] The fault diagnosis module analyzes the data fluctuation coefficient, signal packet loss rate and autocorrelation function, matches the preset fault feature library, and realizes the identification and location of sensor zero drift, range attenuation and communication interruption faults.

[0021] Furthermore, in the data processing and transmission module, the three-stage data cleaning pipeline includes time-domain filtering based on a cascade structure of median filtering and mean filtering, wavelet transform frequency-domain filtering based on Meyer wavelet basis functions, and anomaly detection based on the isolation forest algorithm. A multi-protocol fusion transmission architecture is constructed, and a hybrid scheme combining discrete cosine transform with run-length coding and Huffman coding is used to achieve lightweight data compression.

[0022] Among them, the time domain filtering adopts the cascade structure of median filtering and mean filtering to suppress Gaussian noise;

[0023] Frequency domain filtering performs multi-resolution analysis based on Meyer wavelet basis functions to eliminate high-frequency interference;

[0024] The isolation forest algorithm constructs no less than 10 isolated trees and uses a voting mechanism to determine the anomaly score to remove outliers.

[0025] Furthermore, in the data receiving and storage module, a triple data verification mechanism includes configuring CRC-32 verification, SHA-256 hash fingerprint comparison, and blockchain evidence verification, and implementing a hot and cold data tiered storage strategy based on data access frequency, update time, and business importance;

[0026] The distributed database constructs a three-dimensional index of geographic coordinates, timestamps, and sensor types based on the Hilbert space-filling curve;

[0027] The hot and cold data tiered storage strategy uses a data migration agent service to store the top 20% of accessed data on SSDs, with the remainder stored on HDDs or tapes.

[0028] Remote disaster recovery is based on the Paxos consensus algorithm.

[0029] Furthermore, the data analysis and decision-making module includes a real-time anomaly detection submodule, a time series prediction submodule, a spatial analysis submodule and a multimodal data fusion submodule;

[0030] Among them, the real-time anomaly detection submodule uses the 3σ criterion for preliminary screening, combines the Bayesian dynamic model to update the anomaly probability, and locates the anomaly points through the isolation forest algorithm;

[0031] The time series prediction submodule uses an LSTM-Transformer hybrid neural network. The Transformer encoder contains a 6-layer multi-head attention mechanism, and the LSTM unit contains 128 memory units.

[0032] The spatial analysis submodule constructs a three-dimensional stress field model based on the Kriging interpolation method and generates a stress distribution cloud map with a resolution of 5m;

[0033] The multimodal data fusion submodule realizes collaborative analysis of heterogeneous data through data preprocessing, feature alignment and deep fusion.

[0034] Furthermore, in the multimodal data fusion submodule,

[0035] The data preprocessing unit performs three-layer wavelet packet decomposition on the stress data and performs semantic segmentation on the image data based on the U-Net model;

[0036] The feature alignment unit is based on a spatiotemporal hash table and uses SIFT feature matching and RANSAC algorithm to achieve precise alignment with a timestamp error of ±1ms and a spatial coordinate error of ±5cm.

[0037] The deep fusion unit uses a multi-scale 3D convolutional neural network with 4 convolutional layers, introduces a multi-head attention mechanism, and optimizes model parameters through end-to-end training.

[0038] Furthermore, the graded warning thresholds are set according to the GB50086-2015 standard, with a yellow warning corresponding to a stress value reaching 80% of the design limit, an orange warning corresponding to 90%, and a red warning corresponding to 100%;

[0039] Supports emails based on SMTP protocol, SMS based on HTTP protocol, and on-site sound and light multi-channel early warning notifications based on Modbus protocol;

[0040] The maintenance decision model adopts a genetic algorithm with safety risk, maintenance cost, and construction period as the objective function, the number of iterations ≥ 100 times, and the population size 50 to solve the optimal resource allocation plan.

[0041] Furthermore, in the user interaction and management module, the multi-terminal adaptation interface is based on WebGL technology and BIM models, achieving 1:1 scale 3D real-time rendering, supporting historical data backtracking with a timeline accuracy of 1 minute and multi-source data superposition;

[0042] Hierarchical permission management is based on the RBAC model, with three roles: administrator, engineer, and observer, corresponding to different functional operations, data access, and system configuration permissions;

[0043] System parameter management supports online adjustment of TCP / IP protocol parameters and database connection parameters.

[0044] Furthermore, the system also includes:

[0045] The security and collaboration module, deployed on field gateway devices, implements edge data preprocessing and collaborative training based on the federated learning framework and Secure Aggregation protocol. It uses the AES-256-GCM symmetric encryption algorithm combined with the ECDH-P256 key exchange protocol and blockchain evidence storage technology based on Hyperledger Fabric. The key is rotated every 30 days. The intrusion detection system combines feature matching based on the Snort rule engine and behavioral analysis technology based on long short-term memory networks.

[0046] In the edge computing architecture of the security and collaboration module, the data filtering layer filters invalid data based on regular expression matching and decision tree rules;

[0047] The feature extraction layer deploys a lightweight deep learning model with a parameter size of no more than 1MB;

[0048] The decision execution layer uses a finite state machine to implement decision logic;

[0049] The federated learning framework aggregates model parameters through the Secure Aggregation protocol, and blockchain evidence storage technology records data operation logs based on the Hyperledger Fabric platform.

[0050] Furthermore, the system also includes:

[0051] The system interaction bus adopts a publish-subscribe message bus architecture, supports asynchronous communication between modules, and the message format includes a 16-bit sensor ID, a 32-bit stress value, and a 64-bit timestamp field;

[0052] Among them, the system interaction bus adopts binary protocol to transmit data, and the message format contains 16-bit sensor ID, 32-bit stress value, and 64-bit timestamp field;

[0053] The data analysis module and the decision-making module interact through a RESTful API that complies with the OpenAPI specification, uses JSON format to transmit data, and implements remote procedure calls based on the gRPC framework. The interface response time does not exceed 500ms.

[0054] The logs of each module are connected to the ELK architecture through the Logstash-Beats collector.

[0055] (3) Beneficial effects

[0056] The present invention provides a wireless transmission anchor rod multi-point stress sensor system, which has the following beneficial effects:

[0057] Multi-protocol adaptive communication and quantum encryption. The multi-protocol adaptive communication stack is based on an improved proximal strategy optimization algorithm. It can complete intelligent switching of communication protocols in a very short time according to the actual environment. In scenarios such as mountain tunnels with complex and changeable signals, it effectively avoids the problem of traditional single-protocol transmission being easily interrupted and ensures the continuity of data transmission. At the same time, the integrated quantum key distribution unit uses the BB84 protocol and combines it with a two-factor encryption system to significantly improve data transmission security. In monitoring scenarios with extremely high data security requirements, such as defense projects and nuclear power plant areas, it can effectively resist quantum attacks and eliminate the risk of data leakage.

[0058] Intelligent sensor management and sensor self-calibration procedures comply with the IEEE1451 standard and utilize least squares fitting to greatly simplify the calibration process, significantly shorten calibration time, and achieve higher calibration accuracy. This allows for efficient calibration in projects involving a large number of sensors, such as large-scale water conservancy projects. A fault diagnosis module, combining a local outlier factor algorithm with an autoregressive sliding average model, monitors sensor status in real time and accurately identifies faults such as zero drift and range attenuation. This prevents data distortion caused by sensor failures in complex environment monitoring, such as mine slopes.

[0059] Three-level data cleaning and spatiotemporal alignment: The three-level data cleaning pipeline effectively removes all types of noise and abnormal data through time-domain and frequency-domain filtering and isolation forest algorithm anomaly detection. This significantly improves data quality in scenarios such as coastal subway tunnels, which are subject to strong vibration and electromagnetic interference. Spatiotemporal alignment technology based on Beidou satellite timing and the seven-parameter Bursa-Wolf model achieves precise time synchronization and spatial positioning, enabling accurate matching of multi-source data and providing strong support for risk analysis.

[0060] Multimodal data fusion, multimodal data fusion technology comprehensively uses the U-Net model, SIFT feature matching and multi-scale 3D convolutional neural network, which can effectively integrate multi-source heterogeneous data such as anchor stress, geological radar images, and surrounding building settlement. In scenarios such as urban deep foundation pit monitoring, compared with traditional systems, it can detect potential instability risks of support structures in advance and take timely measures to avoid accidents.

[0061] Digital twins drive decision-making. The ABAQUS-based three-dimensional digital twin model, combined with the unscented Kalman filter algorithm, can calibrate model parameters in real time based on real-time monitoring data. This allows accurate prediction of surrounding rock stress changes under complex conditions, such as high-speed railway tunnels crossing fault fracture zones, effectively reducing false alarms and missed alarms and ensuring smooth construction progress. A wide range of maintenance plans are generated based on simulations of various extreme conditions. In scenarios such as maintaining old slopes, reinforcement plans can be planned in advance to avoid disasters.

[0062] An intelligent decision-making model, based on relevant standards, uses a genetic algorithm for multi-objective optimization. In projects such as highway slope maintenance, it can comprehensively consider factors such as safety, cost, and construction period, achieve rational allocation of maintenance resources, and reduce maintenance costs and shorten construction periods while ensuring safety.

[0063] A self-organizing edge network based on the Raft algorithm offers fast failover and efficient load balancing. In projects such as large-scale bridge health monitoring, when an edge gateway fails, the system can quickly elect a master node and take over tasks, ensuring uninterrupted monitoring services. Compared to traditional fixed architectures, system stability is greatly improved.

[0064] A trusted security system uses the AES-256-GCM algorithm with quantum key assisted updates for data encryption, combined with a blockchain evidence storage system based on Hyperledger Fabric. This ensures data integrity and authenticity during transmission and storage, providing tamper-proof and reliable data evidence for project acceptance and responsibility determination.

[0065] 3D visualization and interaction: A 1:1 precision 3D visualization system for BIM models based on WebGL technology, supporting VR / AR interaction. Engineers can view anchor stress distribution through an immersive experience, locating problems more quickly and accurately than traditional 2D charts.

[0066] The intelligent management platform, with permission control based on the RBAC model and a log audit system based on the ELK architecture, enables fine-grained permission management and operation traceability for different roles. The Prometheus-Grafana performance monitoring platform monitors the system's operating status in real time, significantly shortening troubleshooting time and ensuring stable system operation.

[0067] In summary, the present invention comprehensively improves the performance of the anchor stress monitoring system through the coordinated application of innovative technical solutions, provides a reliable and intelligent solution for geotechnical engineering safety monitoring, and has important application value and broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flowchart of the overall structure of the present invention. DETAILED DESCRIPTION

[0069] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0070] In geotechnical engineering projects such as tunnels, slopes, and deep foundation pits, the stress state of anchor support structures is a core indicator for assessing project stability. Traditional monitoring technologies rely on single wired / wireless communication protocols and manual analysis, exposing three major technical bottlenecks: First, data transmission reliability is insufficient in complex geological environments, with packet loss rates as high as 15%-20% in typical scenarios, resulting in the loss of critical data. Second, the accuracy of multi-sensor spatiotemporal alignment is limited, with time synchronization errors exceeding 10μs and spatial positioning deviations greater than 20cm, making it difficult to support multi-source data correlation analysis. Third, decision support relies on fixed threshold warnings, which cannot adapt to dynamic changes in surrounding rock stress, resulting in omission and false alarm rates as high as 12% and 18%, respectively. With the increasing demand for intelligent monitoring of major projects under the "new infrastructure" framework, the development of new monitoring systems with environmental adaptability, data accuracy, and decision-making intelligence has become an urgent industry need.

[0071] The proposed wireless transmission anchor multi-point stress sensor system breaks through the technical limitations of traditional monitoring systems through the deep collaboration of edge computing and cloud intelligence, and innovatively constructs core modules such as multi-protocol adaptive communication, intelligent sensor management, multimodal data fusion and digital twin-driven decision-making, forming a full-chain technology system from data acquisition and processing to decision-making. Engineering application verification shows that the system significantly improves the monitoring accuracy and decision-making efficiency in complex environments, providing an innovative technical path for the safe operation and maintenance of geotechnical engineering.

[0072] In geotechnical engineering, anchor support is a key means to ensure structural stability. Traditional monitoring methods rely on manual inspections and wired sensors, and have problems such as low communication reliability, insufficient data processing accuracy, and weak intelligent decision-making capabilities. For example, the data transmission failure rate of a single communication protocol in a complex environment is as high as 15% to 20%; the spatiotemporal alignment accuracy of multiple sensors is poor, the time synchronization error exceeds 10 microseconds, and the spatial positioning deviation is greater than 20 centimeters; the warning threshold is fixed and cannot adapt to dynamic working conditions. These defects seriously restrict the intelligent development of geotechnical engineering monitoring.

[0073] Example: See Figure 1 This embodiment provides a wireless transmission anchor rod multi-point stress sensor system, the specific contents of the system are as follows:

[0074] 1. Data perception and collection

[0075] 1.1 Multi-protocol adaptive communication stack:

[0076] A protocol selection model is constructed based on the Markov decision process (MDP). The state space S = {s1, s2, s3, s4}, where s1 is the signal strength, s2 is the packet loss rate, s3 is the available bandwidth, and s4 is the environmental parameter stability index; the action space A = {a1, a2, a3, a4}, corresponding to ZigBee, LoRa, 4G, and 5G protocols, and the action value function Q(s, a) is iteratively updated through the Q-learning algorithm: Q(s t ,a t )←Q(s t ,a t )+α(r t +γmax a′ Q(s t+1 ,a′)-Q(s t ,a t ));

[0077] Among them, α is the learning rate, γ is the discount factor, and r t For instant rewards, through the multi-dimensional reward function r t =0.4s 1t +0.3(1-s 2t )+0.2s 3t +0.1(1 / s 4t ) calculation, where s 1t 、s 2t 、s 3t 、s 4tThey represent the signal strength at time t (the value range is [-120, 0] dBm, with larger values ​​indicating stronger signals), packet loss rate (the value range is [0, 100]%, with lower values ​​indicating more stable transmission), available bandwidth (the larger the bandwidth, the faster the transmission speed, in Mbps), and environmental parameter stability index (the closer to 1, the more suitable the environment for data transmission), ensuring that the protocol switching delay is controlled within 200 ms.

[0078] For example, during monitoring in a mountain tunnel, when the LoRa protocol signal strength s1 dropped to -120dBm and the packet loss rate s2 exceeded 5%, the (Q) value of each protocol was calculated according to the above formula, and the system triggered protocol switching, completing the migration to the 4G network within 180ms to ensure stable data transmission.

[0079] At a certain moment, the state data collected by the sensor is: signal strength s1 = -80dBm, packet loss rate s2 = 3%, available bandwidth s3 = 5Mbps, power consumption s4 = 0.8. Substituting these data into the reward function, we can get: r t =0.4×(-80)+0.3×(1-3)+0.2×5+0.1×(1 / 0.8)=-31.475.

[0080] The system selects the protocol with the largest reward value for data transmission based on the reward values ​​under different protocols. If the reward value calculated by the LoRa protocol is higher than that of other protocols, the system automatically switches to the LoRa protocol within 200ms to ensure uninterrupted data transmission.

[0081] 1.2 Quantum Enhanced Secure Communication

[0082] The BB84 protocol quantum key distribution unit is integrated. The sender prepares single photons in four polarization states, and the receiver randomly selects a measurement basis. The two parties compare the measurement basis information to filter the original key. After error correction and privacy amplification, a 256-bit true random quantum key is generated. It is transmitted in parallel with the classical channel through wavelength division multiplexing. The key negotiation delay is ≤50ms. The anti-quantum attack capability complies with the GM / T0092-2021 standard, ensuring the security of data transmission in high-security scenarios.

[0083] Analysis: Quantum key distribution is based on the BB84 protocol and uses the polarization state of photons to transmit keys. The sender randomly selects four polarization states (horizontal, vertical, +45°, -45°) to prepare single photons, and the receiver randomly selects two measurement bases (horizontal / vertical basis, +45° / -45° basis) for measurement. The two parties publicly compare part of the measurement information, screen out consistent measurement results as the original key, discard the inconsistent parts, and then generate the final secure key through error correction and privacy amplification. This mechanism uses the principle of quantum non-cloning to ensure key security, and the anti-quantum attack capability meets the GM / T0092-2021 standard.

[0084] 1.3. Intelligent sensor management:

[0085] The sensor self-calibration procedure follows the IEEE1451 standard and collects the sensor output value y by injecting five sets of step wave signals [0V, 2V, 4V, 6V, 8V]. i , use the least squares method to fit y = kx + b, where k is the sensitivity coefficient and b is the zero offset, by minimizing the error function Determine the sensitivity coefficient and zero offset to improve calibration efficiency and accuracy. Take the partial derivatives of E with respect to k and b respectively and set them to 0, and we can get:

[0086]

[0087] Solving the above system of equations, we can get the values ​​of k and b.

[0088] Example: In a new tunnel project, the anchor stress sensor numbered T-001 was calibrated, and the five sets of data collected are as follows:

[0089]

[0090]

[0091] The fault diagnosis module combines the local outlier factor (LOF) algorithm and the autoregressive moving average (ARMA) model. The LOF algorithm calculates the local reachable density of the data point x. and local outlier factor LOF(p) to identify anomalies.

[0092] A 5-point median filter removes impulse noise, and a 10-point mean filter suppresses Gaussian noise, improving the signal-to-noise ratio by more than 15dB. A 3-layer wavelet decomposition based on the Meyer wavelet basis retains the 0-10Hz effective frequency band and filters out mechanical vibration (>20Hz) and electromagnetic interference (>50Hz).

[0093] The isolation forest algorithm is used to construct 20 isolated trees, and the average path length of the data points is calculated. The outlier threshold is set to 3 times the standard deviation. The data point set is D. For the data point x∈D, its path length in the i-th isolated tree is h i (x), the average path length is

[0094]

[0095] Among them, N(x) is the neighborhood point set of data point x, d(x,y) is the distance between x and y.

[0096] When LOF(p)>3, fault detection is triggered, and the data point p is determined to be an outlier. The ARMA model is combined with the signal trend prediction to realize multi-dimensional diagnosis of sensor faults. The LOF algorithm calculates the local reachable density and local outlier factor of the data point to determine the abnormality, and the ARMA model predicts the data trend. The two are combined to diagnose sensor zero drift, range attenuation and other faults.

[0097] Example: In a bridge construction monitoring, anomaly detection is performed on anchor stress data. A data set D containing 100 data points is selected and 20 isolated trees are constructed. For a data point x0, the path lengths in each isolated tree are calculated to be h1(x0)=3, h2(x0)=4, ..., h 20 (x0)=5, then the average path length Assume that its neighborhood point set N k (p) contains 10 points, and the sum of reach-dist(x0,y) is 50, then If LOF(x1)=3.5, and LOF(x1)=3.5, and LOF(x1)=3.5, then x1 is judged to be an outlier and needs further investigation and processing.

[0098] 2. Data processing and transmission module

[0099] 2.1. Three-level data cleaning pipeline:

[0100] Time domain filtering adopts a cascade structure of median filtering and mean filtering. Median filtering removes impulse noise by sorting the data in the window and taking the median. Mean filtering calculates the average value of the data in the window to suppress Gaussian noise, thereby improving the signal-to-noise ratio by more than 15dB.

[0101] Frequency domain filtering performs three-layer wavelet decomposition based on Meyer wavelet basis, decomposing the signal x(t) into approximate components A j and detail component D j :

[0102]

[0103] By selecting the appropriate number of decomposition layers and thresholds, the effective signal frequency band of 0-10Hz is retained and high-frequency vibration interference is filtered out.

[0104] The isolation forest algorithm is used for anomaly detection. 20 isolated trees are constructed and the degree of anomaly is evaluated by calculating the average path length h(x) from the data point to the root node. The outlier determination threshold is set to 3 times the standard deviation of the average path length.

[0105] 2.2, Spatiotemporal Alignment Technology:

[0106] Time synchronization is based on the 1PPS (pulse per second) signal of the BeiDou satellite B1 frequency point. The sensor network time synchronization is achieved through the NTP protocol with a synchronization accuracy of ±1 microsecond. The spatial correction adopts the seven-parameter Bursa-Wolf model (dx, dy, dz, w x ,w y ,w z ,k), the sensor coordinate system is converted to the engineering coordinate system by the following formula:

[0107]

[0108] Among them, (X old ,Y old ,Z old ) is the sensor coordinate, (X new ,Y new ,Z new ) is the reference coordinate, dx, dy, dz are the translation parameters in three directions, w x ,w y ,w z are the rotation parameters in three directions, k is the scale parameter, and the seven parameters are optimized by the least squares matching algorithm to make the spatial positioning error ≤ 5 cm.

[0109] For example, in mountain slope monitoring, the coordinates of an anchor sensor in the original coordinate system are known to be (100, 200, 300) m, and the seven parameters are dx = 0.1 m, dy = 0.2 m, dz = 0.3 m, w x =0.001rad, w y =0.002rad,w z =0.003rad, k=0.0001, and substituting them into the above formula, we can obtain the coordinates of the anchor sensor as (100.11, 200.22, 300.33)m, and the spatial positioning error is within the range of ±5cm.

[0110] 3. Data analysis and decision-making module

[0111] 3.1 Multimodal Data Fusion

[0112] The data preprocessing layer performs three-layer wavelet packet decomposition on the stress data to obtain eight sub-bands; the geological radar image is semantically segmented using the U-Net model, and the drone video is detected using the YOLOv5m model;

[0113] The feature alignment layer is based on the spatiotemporal hash table H(t,x,y,z), uses the SIFT feature matching algorithm to find points with the same name, and combines it with the RANSAC algorithm to eliminate mismatched points to improve the precise alignment of timestamps and spatial coordinates.

[0114] The deep fusion layer uses a multi-scale 3D convolutional neural network (4 convolution blocks, each containing a 3×3×3 kernel) to extract spatiotemporal features through the convolution operation y=f(W*x+b), where W is the convolution kernel, x is the input feature map, b is the bias, and f is the activation function. A multi-head attention mechanism is introduced to weightedly fuse features of different modalities and output a 128-dimensional fused feature vector.

[0115] 3.2 Digital Twins Drive Decision-Making

[0116] A three-dimensional finite element model was established based on ABAQUS. The model was divided into more than 100,000 tetrahedral elements through discretization. Physical quantities such as the anchor axial force and mortar bond strength were calculated according to the elastic mechanics equations:

[0117]

[0118] Among them, σ is the stress tensor, ε is the strain tensor, b is the body force, t is the surface force, and u is the displacement.

[0119] The unscented Kalman filter (UKF) algorithm is used to update the model parameters every hour, by selecting the Sigma point set χ (i) Approximate state distribution:

[0120]

[0121] in, is the state estimate, P is the covariance matrix, n is the state dimension, and λ is the scaling parameter. The model parameters are adjusted through the prediction and update steps to control the calibration error.

[0122] The Unscented Kalman Filter (UKF) algorithm is used to update the model parameters every hour, with a calibration error of ≤3%. The system state equation is x k =f(x k-1 )+w k-1 , the observation equation is z k =h(x k )+v k , where x k is the system state vector, f(x) is the state transfer function, w k is the process noise, z k is the observation vector, h(x) is the observation function, v k To observe the noise, the UKF algorithm selects a set of Sigma points To approximate the state distribution, after time update and measurement update steps, the state estimate is calculated and covariance P k|k .

[0123] It supports simulations of eight extreme working conditions, including blasting vibration and groundwater infiltration, and automatically generates more than 20 maintenance plans based on the simulation results. For example, in mine slope monitoring, it simulates the redistribution of surrounding rock stress under rainfall conditions and predicts the location of potential slip surfaces on the slope.

[0124] Analysis: The digital twin model is constructed based on finite element analysis. The anchor support structure is divided into a large number of units through discretization, and the stress and strain distribution of each unit is obtained by solving the elastic mechanics equations. The UKF algorithm selects Sigma point sets to approximate the state distribution, which can more accurately handle nonlinear systems. The model parameters are updated every hour to ensure that the simulation results are highly consistent with the actual engineering conditions. Taking a large-scale water conservancy project as an example, by simulating the mechanical behavior of anchors under different working conditions, potential risks can be predicted in advance, providing a scientific basis for project maintenance.

[0125] 4. Edge computing and security module

[0126] 4.1 Self-organizing edge network:

[0127] The Raft algorithm is used to implement self-organization of edge nodes. Node states are divided into leaders, followers, and candidates. The leader maintains cluster consistency through heartbeat packets (sending interval ≤ 200ms).

[0128] Layered architecture: data filtering layer (rule engine filters 90% of invalid data), feature extraction layer (TinyML model, parameter size ≤ 1MB, latency ≤ 100ms), decision execution layer (finite state machine, 15 transition rules);

[0129] Distributed collaboration: Based on the Raft algorithm, edge node self-organization is achieved, with failover time ≤ 500ms, load balancing ≥ 0.95, and support for dynamic expansion to 100+ nodes

[0130] When the leader fails, the candidate initiates an election. The candidate that receives votes from more than half (≥50% + 1) of the nodes becomes the new leader. The failover time is ≤500ms. The consistent hashing algorithm is used to distribute computing tasks. The hash value range that the node is responsible for is determined by the following formula:

[0131] h(key)=hash(key)mod N;

[0132] Among them, h(key) is the hash value, N is the number of nodes, and it adjusts the load balancing of edge nodes.

[0133] Analysis: The Raft algorithm maintains cluster status through heartbeat detection (period ≤ 200ms). When the leader node fails, the follower node waits for the election timeout (default 150-300ms) before becoming a candidate and sending voting requests to other nodes. The candidate that receives votes from more than half of the nodes becomes the new leader. The entire election process is completed within 500ms, ensuring uninterrupted monitoring services. The consistent hashing algorithm achieves load balancing by mapping data and nodes to the same hash space, effectively improving the stability of the system in complex environments.

[0134] 4.2 Trusted Security System

[0135] Data encryption uses the AES-256-GCM algorithm to encrypt data payloads, and session keys are exchanged through the ECDH-P256 protocol. Quantum keys are assisted in updates, and the key update cycle is 15 minutes. Blockchain evidence storage is based on the Hyperledger Fabric consortium chain. Each block contains 50 transactions, and the SHA-3 hash algorithm is used to generate fingerprints, effectively limiting the probability of data tampering.

[0136] 5. User interaction and management module

[0137] WebGL technology is used to achieve 1:1 precision 3D real-time rendering of BIM models. An octree structure is used to manage 1,000+ anchor models. The rendering frame rate is calculated using the following formula:

[0138]

[0139] Among them, t render For single rendering time, ensure the frame rate ≥ 30FPS, support VR / AR interaction mode, and gesture recognition delay ≤ 200ms.

[0140] Based on the RBAC model, three roles are set: administrator, engineer, and observer, and fine-grained management and control are achieved through the permission matrix R×P (role×permission).

[0141] 6. System deployment and implementation

[0142] 6.1. Hardware equipment selection and installation:

[0143] Sensor deployment: In a subway tunnel project, a vibrating string anchor stress sensor was installed every 10 meters, with a total of 500 sensors deployed. When installing the sensor, ensure that it is tightly connected to the anchor and fix it with a special clamp to prevent looseness from affecting measurement accuracy.

[0144] For edge gateway settings, one edge gateway is configured for every 100 sensors. A gateway device with an industrial-grade ARM Cortex-A72 processor, 2GB RAM, and 16GB Flash is selected. It is installed in a waterproof chassis on the side wall of the tunnel and connected to the sensors via optical fiber to ensure stable communication.

[0145] Quantum communication equipment is deployed, and quantum key distribution equipment is deployed in key areas of the tunnel. The quantum transmitter and receiver are installed between the equipment at both ends of the tunnel, connected by single-mode optical fiber, and optical path calibration is performed to ensure minimal quantum signal transmission loss.

[0146] Cloud server configuration: Building a distributed server cluster on Alibaba Cloud. Each node is equipped with an Intel Xeon Gold 6240 processor, 128GB of memory, and 2TB of SSD storage. The CockroachDB database and Kubernetes container management platform are installed, along with a network load balancer to achieve high-concurrency data processing.

[0147] 6.2 Software system installation and configuration:

[0148] Edge software deployment: build an embedded Linux system based on Yocto, burn it to the edge gateway, install the data perception and acquisition module, data processing and transmission module, edge computing module, security and collaboration module, configure the adaptive communication protocol stack, set the working mode of the quantum key distribution module, and adjust the data filtering rules and feature extraction model parameters of the edge computing unit.

[0149] Cloud software deployment, using Docker containers to deploy data analysis and decision-making modules, user interaction and management modules, data reception and storage modules, container orchestration through Kubernetes, configuration of the distributed database's three-dimensional index structure and hot and cold data tiered storage strategy, setting training parameters for data analysis models, and configuration of the user rights management system.

[0150] 6.3 Data Collection and Processing Implementation

[0151] During the data collection process, the data perception and acquisition module controls the sensor to collect stress data at a frequency of 10Hz, and simultaneously obtains information such as the sensor's battery level and signal strength. The adaptive communication protocol stack monitors the communication status in real time. When the 4G network signal strength is lower than -100dBm, the (Q) value of each protocol is calculated according to the MDP-Q-learning algorithm, and the module automatically switches to the LoRa protocol for data transmission.

[0152] During the data processing process, the collected data first enters a three-level data cleaning pipeline. Time domain filtering removes pulse noise caused by tunnel construction vibration, frequency domain filtering filters out 50Hz power frequency interference, and the isolation forest algorithm detects and eliminates abnormal data points. Then, time and space alignment is performed, and time synchronization is achieved using the Beidou timing signal. The spatial coordinate conversion is completed through the seven-parameter Bursa-Wolf model. Finally, the feature engineering module extracts 15 characteristic parameters such as stress gradient standard deviation and fluctuation entropy value, constructs feature vectors, and transmits them to the cloud.

[0153] 6.4. Multimodal data fusion analysis. In a certain slope monitoring project, stress data are fused and analyzed with geological radar images. The stress data are decomposed by three layers of wavelet packets and then combined with the geological radar images segmented by the U-Net model through a spatiotemporal hash table.

[0154] In the application, the several formulas involved are all calculated by taking their numerical values ​​after removing the dimensions, and the formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent real situation. Some coefficients or weights in the formulas are set by technical personnel in this field according to actual conditions, so they will not be elaborated here.

[0155] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0156] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0157] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A wireless transmission anchor rod multi-point stress sensor system, characterized in that: The system includes: The data sensing and acquisition module establishes a two-way communication link with the anchor stress sensor based on an adaptive communication protocol stack, and uses a strategy combining polling and event triggering to collect data in real time. The protocol stack supports dynamic switching of multiple protocols. The data processing and transmission module builds a three-level data cleaning pipeline. Based on the timestamp signal provided by the Beidou satellite timing system, it combines the seven-parameter Bursa-Wolf transformation model with the least squares matching algorithm to perform spatiotemporal alignment, extract multi-dimensional feature parameters, build a multi-protocol fusion transmission architecture, and perform lightweight data compression. The data reception and storage module is deployed in the cloud or on a local server, equipped with a triple data verification mechanism, a distributed database architecture with a three-dimensional index structure based on the Hilbert space-filling curve, a hierarchical storage strategy, and remote disaster recovery with real-time log synchronization between the primary and standby databases based on the Paxos consensus algorithm. The data analysis and decision-making module connects to the engineering knowledge base to build a hierarchical warning and maintenance decision-making model based on multi-objective optimization; The user interaction and management module provides a multi-terminal adaptation interface, realizes 3D real-time rendering based on BIM model and WebGL technology, and integrates hierarchical permission management and system parameter management modules based on the RBAC model.

2. A wireless transmission anchor rod multi-point stress sensor system according to claim 1, characterized in that: In the data sensing and acquisition module, the adaptive communication protocol stack uses a protocol selection algorithm based on the Markov decision process, constructs a Q value matrix based on signal strength, packet loss rate, and available bandwidth, performs protocol switching through dynamic programming, and uses a strategy combining polling and event triggering to collect data in real time, including stress values ​​and three-dimensional stress components. The protocol stack supports multiple protocols including ZigBee and LoRa, and has a built-in sensor self-calibration program compliant with the IEEE1451 standard and a fault diagnosis module based on the local outlier factor algorithm and the autoregressive moving average model; The sensor self-calibration program injects a step wave excitation signal, collects no less than five sets of signal data with different excitation intensities, and uses the least squares curve fitting method to determine the calibration parameters by minimizing the sum of squared errors; The fault diagnosis module analyzes the data fluctuation coefficient, signal packet loss rate and autocorrelation function, matches the preset fault feature library, and realizes the identification and location of sensor zero drift, range attenuation and communication interruption faults.

3. The wireless transmission anchor rod multi-point stress sensor system according to claim 1, characterized in that: In the data processing and transmission module, the three-stage data cleaning pipeline includes time-domain filtering based on a cascaded structure of median filtering and mean filtering, wavelet transform frequency-domain filtering based on Meyer wavelet basis functions, and anomaly detection based on the isolation forest algorithm. A multi-protocol fusion transmission architecture is constructed, and a hybrid scheme combining discrete cosine transform with run-length coding and Huffman coding is used to achieve lightweight data compression. Among them, the time domain filtering adopts the cascade structure of median filtering and mean filtering to suppress Gaussian noise; Frequency domain filtering performs multi-resolution analysis based on Meyer wavelet basis functions to eliminate high-frequency interference; The isolation forest algorithm constructs no less than 10 isolated trees and uses a voting mechanism to determine the anomaly score to remove outliers.

4. The wireless transmission anchor rod multi-point stress sensor system according to claim 1, characterized in that: In the data receiving and storage module, a triple data verification mechanism includes configuring CRC-32 verification, SHA-256 hash fingerprint comparison, and blockchain evidence verification, and implementing a hot and cold data tiered storage strategy based on data access frequency, update time, and business importance; The distributed database constructs a three-dimensional index of geographic coordinates, timestamps, and sensor types based on the Hilbert space-filling curve; The hot and cold data tiered storage strategy uses a data migration agent service to store the top 20% of accessed data on SSDs, with the remainder stored on HDDs or tapes. Remote disaster recovery is based on the Paxos consensus algorithm.

5. The wireless transmission anchor rod multi-point stress sensor system according to claim 1, characterized in that: The data analysis and decision-making module includes a real-time anomaly detection submodule, a time series prediction submodule, a spatial analysis submodule and a multimodal data fusion submodule; Among them, the real-time anomaly detection submodule uses the 3σ criterion for preliminary screening, combines the Bayesian dynamic model to update the anomaly probability, and locates the anomaly points through the isolation forest algorithm; The time series prediction submodule uses an LSTM-Transformer hybrid neural network. The Transformer encoder contains a 6-layer multi-head attention mechanism, and the LSTM unit contains 128 memory units. The spatial analysis submodule constructs a three-dimensional stress field model based on the Kriging interpolation method and generates a stress distribution cloud map with a resolution of 5m; The multimodal data fusion submodule realizes collaborative analysis of heterogeneous data through data preprocessing, feature alignment and deep fusion.

6. The wireless transmission anchor rod multi-point stress sensor system according to claim 5, characterized in that: In the multimodal data fusion submodule, The data preprocessing unit performs three-layer wavelet packet decomposition on the stress data and performs semantic segmentation on the image data based on the U-Net model; The feature alignment unit is based on a spatiotemporal hash table and uses SIFT feature matching and RANSAC algorithm to achieve precise alignment with a timestamp error of ±1ms and a spatial coordinate error of ±5cm. The deep fusion unit uses a multi-scale 3D convolutional neural network with 4 convolutional layers, introduces a multi-head attention mechanism, and optimizes model parameters through end-to-end training.

7. The wireless transmission anchor rod multi-point stress sensor system according to claim 5, characterized in that: The graded warning thresholds are set according to GB50086-2015 standard. Yellow warning corresponds to stress values ​​reaching 80% of the design limit, orange warning corresponds to 90%, and red warning corresponds to 100%. Supports emails based on SMTP protocol, SMS based on HTTP protocol, and on-site sound and light multi-channel warning notifications based on Modbus protocol; The maintenance decision model adopts a genetic algorithm with safety risk, maintenance cost, and construction period as the objective function, the number of iterations ≥ 100 times, and the population size 50 to solve the optimal resource allocation plan.

8. The wireless transmission anchor rod multi-point stress sensor system according to claim 1, characterized in that: In the user interaction and management module, the multi-terminal adaptation interface is based on WebGL technology and BIM models, achieving 1:1 scale 3D real-time rendering, supporting historical data backtracking with a timeline accuracy of 1 minute and multi-source data overlay; Hierarchical permission management is based on the RBAC model, with three roles: administrator, engineer, and observer, corresponding to different functional operations, data access, and system configuration permissions; System parameter management supports online adjustment of TCP / IP protocol parameters and database connection parameters.

9. The wireless transmission anchor rod multi-point stress sensor system according to claim 1, characterized in that: The system also includes: The security and collaboration module, deployed on field gateway devices, implements edge data preprocessing and collaborative training based on the federated learning framework and Secure Aggregation protocol. It uses the AES-256-GCM symmetric encryption algorithm combined with the ECDH-P256 key exchange protocol and blockchain evidence storage technology based on Hyperledger Fabric. The key is rotated every 30 days. The intrusion detection system combines feature matching based on the Snort rule engine and behavioral analysis technology based on long short-term memory networks. In the edge computing architecture of the security and collaboration module, the data filtering layer filters invalid data based on regular expression matching and decision tree rules; The feature extraction layer deploys a lightweight deep learning model with a parameter size of no more than 1MB; The decision execution layer uses a finite state machine to implement decision logic; The federated learning framework aggregates model parameters through the Secure Aggregation protocol, and blockchain evidence storage technology records data operation logs based on the Hyperledger Fabric platform.

10. The wireless transmission anchor rod multi-point stress sensor system according to claim 1, characterized in that: The system also includes: The system interaction bus adopts a publish-subscribe message bus architecture, supports asynchronous communication between modules, and the message format includes a 16-bit sensor ID, a 32-bit stress value, and a 64-bit timestamp field; Among them, the system interaction bus adopts binary protocol to transmit data, and the message format contains 16-bit sensor ID, 32-bit stress value, and 64-bit timestamp field; The data analysis module and the decision-making module interact through a RESTful API that complies with the OpenAPI specification, uses JSON format to transmit data, and implements remote procedure calls based on the gRPC framework. The interface response time does not exceed 500ms. The logs of each module are connected to the ELK architecture through the Logstash-Beats collector.

Images