A tunnel structure health monitoring and early warning method and system based on multi-source twin data fusion

Abstract

The present invention discloses a tunnel structure health monitoring and early warning system based on multi-source twin data fusion, which relates to the field of tunnel health monitoring technology, and includes four modules: a data acquisition module, a multi-source data fusion module, a health status assessment module, and an early warning module; the data acquisition module can collect equipment status data, tunnel environment data, traffic status data, and various monitoring data of the tunnel structure of the monitoring tunnel; the multi-source data fusion module uses advanced data fusion technology to integrate and process the collected multi-source data; the health status assessment module uses an intelligent assessment algorithm to assess the health status of the tunnel structure based on the fused data and identify potential structural problems; the early warning module uses an early warning algorithm to form an early warning result based on the assessment result. The present invention adopts advanced data fusion methods and early warning algorithms to improve the accuracy and reliability of tunnel health monitoring and early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering operation and maintenance, in particular to the field of tunnel engineering monitoring and safety early warning, and more particularly to application tools for tunnel operation monitoring and early warning implementation methods, specifically a tunnel structure health monitoring and early warning system with multi-source twin data fusion, which is used for real-time monitoring and early warning of the health status of the tunnel structure. Background Art

[0002] During the operation and management stage, tunnels often have cracks, bulges, water leakage and other defects. At present, the discovery of the above defects mainly relies on the naked eye observation of patrol personnel or the use of instruments and equipment for relevant detection (Chongqing Jiaotong University. An intelligent cloud control system in a tunnel based on 5G technology: 202011400906.4 [P]. 2021-04-06.). In the existing technology, tunnel structure health monitoring and early warning systems have been widely used in tunnel engineering to ensure the safety and stability of tunnels (Shanghai Jiaotong University; Shanghai Urban Construction Urban Operation (Group) Co., Ltd. A dynamic early warning method for abnormal data of tunnel health monitoring based on ARIMA model: 202211616781.8 [P]. 2023-05-26). However, the existing monitoring systems usually have the following problems:

[0003] Single data source problem: Most existing monitoring systems rely on a single data source, such as stress, deformation, temperature and other single data, which makes it difficult to fully reflect the overall health status of the tunnel structure.

[0004] Limited data fusion and processing capabilities: Traditional systems lack effective data fusion algorithms when processing multi-source data, resulting in serious data redundancy and noise problems, affecting the accuracy and reliability of monitoring results.

[0005] Imperfect early warning mechanism: The early warning algorithm of the existing system is relatively simple, usually using simple threshold judgments, which cannot accurately assess complex structural risks, and early warning signals are prone to false alarms or omissions.

[0006] In summary, the existing tunnel structure health monitoring system has many shortcomings in terms of data source, data processing, early warning mechanism and data integrity. There is an urgent need for a tunnel structure health monitoring and early warning system that can integrate multiple data sources, adopt advanced data fusion algorithms and intelligent early warning mechanisms, so as to achieve comprehensive monitoring and accurate early warning of tunnel structures and improve the safety and stability of tunnels. Summary of the invention

[0007] The present invention aims to overcome at least one defect of the above-mentioned prior art and provide a tunnel structure health monitoring and early warning system with multi-source twin data fusion, which is used to solve the technical problems in the prior art such as single monitoring data source, limited data processing capability, imperfect early warning mechanism, insufficient data integrity, and inability to accurately warn of complex structure risks.

[0008] The present invention provides a tunnel structure health monitoring and early warning system with multi-source twin data fusion, which is used for real-time monitoring and early warning of the health status of the tunnel structure, including a data acquisition module, a multi-source data fusion module, a health status assessment module, and an early warning module:

[0009] S01, data acquisition module: collects health monitoring data of tunnel structure from multiple data sources;

[0010] S02, multi-source data fusion module: fusing the collected multi-source data;

[0011] S03, health status assessment module: assessing the health status of the tunnel structure based on fusion data;

[0012] S04, early warning module: early warning based on health status assessment results.

[0013] The data acquisition module includes a sensing technology module and a data transmission unit. The sensing technology module is used to collect equipment status data, tunnel environment data, traffic condition data and various monitoring data of the tunnel structure to form multi-source data of the tunnel. The various monitoring data include parameters such as stress, deformation, temperature, humidity, etc. The data acquisition module includes various sensing devices, such as strain gauges, accelerometers, temperature sensors, humidity sensors, displacement sensors, vibration sensors and environmental monitoring sensors, etc.; the data transmission unit is used to transmit the collected data to the data processing center in real time, and has a data preprocessing function. The data preprocessing function includes preliminary filtering and preliminary analysis of the data to ensure the accuracy and reliability of the data.

[0014] The multi-source data fusion module includes a data integration platform module, a data fusion and processing algorithm, and data correction, which are used to integrate and process the collected multi-source tunnel data, thereby improving the accuracy and consistency of the data. The data integration platform is used to uniformly manage and coordinate data from different sensors and monitoring equipment; the data fusion and processing algorithm effectively processes and fuses the multi-source tunnel data through methods such as Kalman filtering, Bayesian reasoning, and principal component analysis, eliminates redundant information and data noise, and improves the reliability and effectiveness of the data. The data correction is used to deal with data missing problems caused by equipment failure or data loss to ensure data integrity.

[0015] The health status assessment module includes an intelligent assessment algorithm, a multi-level health assessment, a historical data comparison and analysis, a detailed assessment report generation and a visual display, and is used to evaluate the health status of the tunnel structure based on the fused data. On the basis of the fused data of the multi-source data fusion module, the health status assessment module uses an intelligent assessment algorithm to evaluate the health status of the tunnel structure and identify potential structural problems. The intelligent assessment algorithm and the multi-level health assessment include a variety of analysis tools and intelligent assessment algorithms, such as finite element analysis, neural network algorithm, fuzzy logic algorithm and expert system, to provide a multi-level health assessment. The historical data comparison and analysis discovers and evaluates the structural change trend by comparing historical data with real-time data; the detailed assessment report generation generates a detailed tunnel health assessment report after analysis, and the tunnel health assessment report includes a structural health score, potential risk locations and possible fault types, and provides corresponding maintenance suggestions for staff; the module also supports the visual display, and intuitively displays the health status and assessment results of the tunnel structure in the form of charts, 3D models, etc., so that relevant personnel can understand and deal with structural problems in a timely manner.

[0016] The early warning module includes a multi-level early warning algorithm, dynamic response strategy recommendations and a multi-channel notification system. According to the health status assessment results, the multi-level early warning algorithm is used to form an early warning result, and relevant personnel are notified through various channels so that corresponding maintenance measures can be taken in a timely manner. The multi-level early warning algorithm includes a threshold algorithm and a probabilistic risk intelligent assessment algorithm to identify different types and levels of structural risks, and conduct graded early warnings, generating early warning signals of different levels according to the degree of risk, to ensure that relevant personnel can take appropriate measures according to the risk level; the multi-channel notification system sends early warning information to relevant personnel through various channels such as text messages, e-mails, mobile applications and Internet of Things devices; the response strategy recommendation function provides specific maintenance and emergency response recommendations based on the different types of early warning signals to help relevant personnel respond to potential structural problems quickly and effectively.

[0017] The sensing technology module is responsible for sensing and collecting key monitoring data of the tunnel structure and the surrounding environment, and is used to flexibly adapt to the needs of collecting structural health monitoring data and geological radar data under different construction and operation environments. The data transmission unit supports wired and wireless network transmission and can be flexibly applied in different construction and operation environments.

[0018] The sensing technology module and the data transmission unit specifically include:

[0019] S01.1. Monitor the stress and strain of the tunnel structure under different load conditions through stress and strain sensors to capture structural deformation information;

[0020] S01.2. Install strain gauges and displacement sensors to record the crack distribution, characteristics and expansion patterns of the tunnel structure, and provide early warning information on potential risks of the structure;

[0021] S01.3. Use temperature sensors to monitor temperature changes inside the tunnel and in the surrounding environment, and assess the impact of temperature on tunnel structural materials and overall stability;

[0022] S01.4. Monitor the humidity level inside and around the tunnel through humidity sensors to detect corrosion and other potential impacts that a humid environment may have on the tunnel structure;

[0023] S01.5. Monitor the displacement and settlement of the tunnel structure through displacement sensors to ensure the overall stability and long-term safety of the structure;

[0024] S01.6. Capture the vibration data of the tunnel structure through vibration sensors and analyze the impact of external environmental factors on the tunnel structure;

[0025] S01.7. The geological radar data acquisition unit uses geological radar technology to conduct non-destructive testing of the rock mass and geological environment around the tunnel, obtain key information such as geological layers, cracks and water flows, and ensure the coordination and safety of the tunnel and the surrounding geological environment;

[0026] The data transmission unit also includes:

[0027] S01.8. The data transmission unit transmits the collected data to the data processing center in real time, supporting wired transmission and wireless transmission;

[0028] S01.9. The data processing center includes preliminary data filtering and preliminary analysis, and performs data preprocessing on the collected real-time data.

[0029] The external environmental factors may include traffic loads, earthquakes, etc. By capturing the vibration data in the tunnel environment and external environmental factors, data support is provided for the formulation of vibration control and protective measures for the tunnel.

[0030] The data transmission unit ensures stable transmission and real-time accessibility of monitoring data in different environments by supporting wired transmission and wireless transmission, adapts to the changing environment of tunnel construction and operation, ensures efficient collection and transmission of data, and provides a reliable data source for subsequent data processing and fusion.

[0031] The preliminary data filtering includes outlier detection and noise filtering;

[0032] S01.9.1. Outlier Detection:

[0033] S01.9.11, Statistical-based methods: For each sensor data (such as stress, temperature), assume that its observations over a period of time follow a normal distribution. Detect outliers by calculating the z-score of each data point:

[0034]

[0035] in x i is the observed value at a certain moment, μ is the mean of historical data, σ is the standard deviation, if z If the value exceeds a certain threshold, the data point is considered to be an outlier.

[0036] S01.9.12, based on the sliding window method: using the sliding window to calculate the local mean and standard deviation , and detect whether the current data significantly deviates from the statistical characteristics within the window:

[0037]

[0038] Among them, N is the window size of the sliding window, t is the sliding window time point, is the observed value at that time point. If the current data x t If the deviation from the mean within the window exceeds a certain threshold, it is marked as an anomaly.

[0039] S01.9.2. The noise filtering includes:

[0040] S01.9.21, Low-pass filtering: By setting a low-pass filter, high-frequency noise in the sensor data is filtered out while retaining the effective components in the signal.

[0041] Let the sampling frequency be f s , the cutoff frequency of the selected filter f c and filter order n , the filter transfer function for:

[0042]

[0043] in w c =2Πf c .

[0044] By applying the discretized transfer function to the sensor data sequence, high frequency noise is eliminated.

[0045] S01.9.22, Median filter: used to remove impulse noise.

[0046] Remove isolated noise points by replacing the central data point with the median value in the data window:

[0047]

[0048] 2 of them k +1 is for window size, is the central data point value after replacement;

[0049] S01.9.3, the data feature extraction includes:

[0050] S01.9.31, Time Domain Feature Extraction: Calculate Mean and standard deviation , and use skewness and kurtosis to capture the asymmetry and peak degree of the data distribution:

[0051]

[0052]

[0053] S01.9.32, Frequency domain feature extraction: Convert time domain data to frequency domain through Fast Fourier Transform FFT to analyze the frequency components of the signal:

[0054]

[0055] in is a discrete-time signal, is the time index, yes The Fourier transform of is the frequency index, is an imaginary unit, satisfying j 2 =-1, N is the sequence Length;

[0056] The power spectral density (PSD) is calculated to identify the main frequency components and their corresponding amplitudes. The PSD can be obtained by calculating the autocorrelation function of the signal and performing a Fourier transform on it:

[0057]

[0058] in R xx (m) is the autocorrelation function of the signal.

[0059] S01.9.4. The trend analysis includes:

[0060] S01.9.41. Linear Regression: Analyze time series trends in sensor data using linear regression , estimate the slope and intercept of the trend to determine whether there is a long-term change in the data:

[0061]

[0062] in is the intercept, is the slope, is the error term, and the parameter is estimated by the least squares method. and .

[0063] S01.9.42 Moving Average: Calculate the moving average of data to smooth short-term fluctuations and show long-term trends.

[0064]

[0065] in, is the moving average at time t, is the window size of the moving average, is the observation at time i, for a given window size , the moving average helps identify potential trend changes and provides a basis for further analysis.

[0066] The information reflecting the real physical state of the tunnel structure in the multi-source data fusion module is a multi-source information set, which includes a real information set of the external environment, a real information set of the structural performance state, and a real information set of the material characteristic state.

[0067] The multi-source data fusion module specifically includes:

[0068] S02.1. The data integration platform module detects a subset of a multi-source information set according to various sensing technologies in the data acquisition module, wherein the multi-source information set includes a real information set of an external environment, a real information set of a structural performance state, and a real information set of a material characteristic state;

[0069] S02.2, reducing the dimension of the multi-source information set by dimensionality reduction technology, eliminating redundant information, simplifying the data set structure, and extracting main features, while applying sufficient constraints based on known tunnel physical information, and outputting an approximate subset of the multi-source information set;

[0070] S02.3. hierarchically process the information of each scale in the approximate subset of the multi-source information set by a multi-scale analysis method, perform precision correction on the information of each scale by combining the spatial resolution, temporal resolution and measurement accuracy of the sensor, and fuse the calibration information at each scale based on the multi-scale reconstruction technology;

[0071] S02.4, estimate the dynamic system in a noisy environment through a recursive algorithm, combine the inherent correlation between the multi-source information, effectively fuse the approximate subset of the multi-source information set, infer the structural state under certain constraints, and dynamically compare it with the real information set detected by the sensing technology module, combine prior knowledge and real-time data, make probability estimation and decision, recursively modify the constraints, and construct complete tunnel structure state information;

[0072] S02.5. The data correction based on the big data integration platform is used to monitor the quality of multi-source data, evaluate and mark the data, detect and correct errors and deviations that may occur in the data collection process of sensors and monitoring equipment, automatically identify abnormal data and make corrections, and handle data missing problems caused by equipment failure or data loss by inferring and completing missing data based on historical big data and maintenance records.

[0073] The data integration platform is characterized by unified management and coordination, real-time data acquisition and efficient data storage management. Through standardized interfaces and protocols, the platform can effectively collect and integrate information from various data sources to ensure seamless connection and unified management of data. The data acquisition platform supports real-time data acquisition and can instantly obtain data from various sensors and monitoring equipment, and perform preliminary processing and storage, ensuring the real-time and accuracy of monitoring data. By adopting distributed storage technology, efficient storage and rapid access to data are ensured, providing powerful data management functions, capable of processing massive data, and supporting the storage and retrieval of historical data.

[0074] In a multi-source information collection, different types of data may be collected at different time and space scales. For example, a temperature and humidity sensor may collect data every hour, while a displacement sensor may collect data every minute. Therefore, it is necessary to perform time alignment and spatial interpolation on the data to ensure that different data sources are unified at the same time and space scales.

[0075] Time alignment: Resample data of different frequencies to the same time node through methods such as linear interpolation or spline interpolation.

[0076] Spatial alignment: If the data sources are located in different locations, such as temperature and humidity sensors and surface deformation measurement equipment, the data can be mapped to the same spatial reference frame through spatial interpolation based on the spatial location and influence range of the sensor layout.

[0077] Dimensionality reduction technology is used to reduce the dimensionality of multi-source information sets to reduce dimensions, eliminate redundant information, and simplify data structures. The core idea is to project high-dimensional data into a low-dimensional subspace while retaining the direction with the largest variance in the data, thereby extracting the most important features.

[0078] S02.2.1. Assume that the multi-source information set is represented as a matrix ,in is the sample size, is the number of features. Each feature is centered and standardized, and the mean is calculated and standard deviation , convert the original data into a standardized matrix :

[0079]

[0080] Calculate the covariance matrix :

[0081]

[0082] Covariance matrix Perform eigenvalue decomposition and obtain the eigenvalue and the corresponding eigenvector The eigenvalue indicates the variance of the principal component, and the eigenvector indicates the direction of the principal component. Sort by eigenvalue from large to small, k The largest eigenvalues ​​and their corresponding eigenvectors form a new low-dimensional subspace.

[0083] The standardized data matrix Projected into this low-dimensional subspace. Assuming that the previous k feature vector , then the new data is expressed as:

[0084]

[0085] in, is a The matrix contains the data after dimensionality reduction. This matrix retains the main features with the largest variance in the original data, thus simplifying the data set while retaining the most important information.

[0086] In the process of dimensionality reduction, physical constraints need to be imposed in combination with known physical information of the tunnel, such as geometric parameters and material properties. The purpose of combining physical constraints is to ensure that the results after dimensionality reduction are not only statistically valid, but also consistent with the structural characteristics and behavior of the tunnel in the actual physical sense.

[0087] S02.2.2. Combined tunnel geometry parameters

[0088] The geometry and dimensions of the tunnel, such as diameter, length, curvature, etc., are known and can be used as important constraints in the dimensionality reduction process. For example:

[0089] Shape constraint: The cross section of a tunnel is usually circular or elliptical, which can be used as a geometric constraint. During the dimensionality reduction process, it can be ensured that the projected data still conforms to this basic shape. For example, if the distribution of a cross-section point cloud data deviates from a circle after dimensionality reduction, it is necessary to adjust it back to the actual tunnel shape through geometric correction.

[0090] Position constraints: The position of the tunnel in space can also be used as a constraint. For example, the axis of the tunnel is usually continuous and smooth, which should be maintained in the reduced-dimensional data. The PCA-projected data can be adjusted to always be distributed along this axis, thus conforming to the overall geometric characteristics of the tunnel.

[0091] S02.2.3. Combining material properties

[0092] Tunnel material properties such as concrete strength, modulus, stress-strain curve, etc. can also be used as constraints:

[0093] Material strength constraints: When evaluating stress and deformation data, if the reduced-dimensional data show that the stress exceeds the yield strength of the material or the deformation exceeds the allowable strain range of the material, these data points can be marked as abnormal or corrected to ensure that the reduced-dimensionality results are within a reasonable material range.

[0094] Material aging and erosion constraints: For long-term tunnels, the materials may change due to aging or erosion from the external environment. These physical properties can be used as constraints in the dimensionality reduction process. For example, if environmental data such as temperature and humidity show that erosion or aging is high in certain areas, the reduced data should reflect the performance changes in these areas.

[0095] S02.2.4. Combining structural mechanics constraints

[0096] Tunnel structures exhibit specific mechanical properties during stress and deformation, such as the correlation between stress distribution and displacement. These properties can be obtained through numerical simulation, structural analysis and field observation, and used as constraints in dimensionality reduction:

[0097] Stress-strain relationship constraint: There is a clear physical relationship between the stress and deformation of the tunnel structure. According to the mechanical model of the material, such as the elastic model, the elastoplastic model, etc., during the dimension reduction process, it can be ensured that the reduced data maintains the relationship between stress and deformation. For example, the principal components after PCA processing should be able to reflect the change pattern of stress and deformation at various positions in the tunnel that conforms to the material mechanics theory.

[0098] Deformation mode constraints: Tunnels may be affected by various factors during operation, such as surface settlement, groundwater changes, and vehicle loads. Based on these known factors and the force mode of the structure, the data can be corrected during the dimensionality reduction process to conform to typical tunnel deformation modes such as longitudinal settlement and lateral contraction. For example, if the data after dimensionality reduction shows deformation that does not conform to physical laws, such as unreasonable and drastic changes, it can be corrected based on the mechanical model of the tunnel.

[0099] S02.2.5. Combining historical data and prior knowledge

[0100] Historical monitoring data and prior knowledge from tunnel operations are also important sources of physical constraints:

[0101] Historical trend constraints: Tunnels accumulate a large amount of historical data during long-term monitoring, which records the deformation, stress changes and other characteristics of the tunnel. By comparing and analyzing these historical data with current data, it is possible to ensure that the dimensionality reduction results are consistent with the long-term trend during the dimensionality reduction process. For example, if historical data shows that a certain area has a slow settlement trend in the past few years, the dimensionality reduction data should be consistent with this trend.

[0102] Prior knowledge constraints: Based on engineering experience and the structural safety specifications for tunnel design, the range of variation of certain data is known. During the dimensionality reduction process, a constraint on the range of data variation can be set so that any principal component data that exceeds this range is marked or corrected.

[0103] These constraints make the dimensionality reduction results more physically meaningful, thereby generating more accurate approximate subsets of multi-source information sets.

[0104] Multiscale analysis methods are used to perform hierarchical processing on approximate subsets of multi-source information sets to capture data features at different scales and fuse information at each scale through multiscale reconstruction techniques.

[0105] S02.3.1. Combination of sensor resolution

[0106] Sensors used in tunnel monitoring usually have different spatial and temporal resolutions. Multiscale analysis methods can process the data in layers according to the resolution of the sensors:

[0107] Spatial resolution: For example, for sensors measuring tunnel lining stress, different spatial resolutions will affect the accuracy of the data. Through multi-scale analysis, data at different resolutions can be hierarchically integrated to unify the scale and improve accuracy.

[0108] Temporal resolution: Different types of sensors (such as accelerometers and displacement sensors) may collect data at different time intervals. Through multi-scale analysis, data at different time scales can be fused to ensure that the structural state is accurately reflected even on a short time scale.

[0109] S02.3.2. Accuracy Calibration

[0110] In the multi-scale analysis process, the accuracy of information at different scales is corrected by combining the measurement accuracy of the sensor and known physical parameters. Specifically, it includes:

[0111] Calibrate small scale based on large scale: By calibrating small scale data on large scale, it is possible to ensure that local information is consistent with the overall trend. For example, in tunnel deformation monitoring, the overall deformation trend can correct the local stress fluctuation data.

[0112] Noise filtering and anomaly detection: Multiscale analysis can also be used to detect and filter data noise at different scales. For example, for vibration data of tunnel linings, wavelet transform can separate high-frequency signals caused by environmental noise or measurement errors and filter them to retain the actual structural response signal.

[0113] S02.3.3. Based on the layered data obtained from multi-scale analysis, the calibration information at each scale can be fused through multi-scale reconstruction technology to restore the overall structural state.

[0114] Multi-scale fusion: By reconstructing low-frequency and high-frequency components at different scales, the complete tunnel structure state containing information at all levels can be obtained. For example, by combining high-frequency components (local fluctuations) with low-frequency components (overall trends), comprehensive state information that conforms to the overall mechanical laws and reflects local anomalies can be reconstructed.

[0115] Fusion formula: Assumptions The weights at each scale are , then the multi-scale reconstruction can be expressed as:

[0116]

[0117] in, In the reconstruction process, the weight and accuracy of each sensor can be combined for weighted fusion to ensure that the final reconstruction result reflects the actual state of the tunnel to the greatest extent.

[0118] By applying multi-scale analysis and reconstruction techniques, physically meaningful structural features and status information can be extracted from complex multi-source information sets, thus providing accurate data support for tunnel health assessment and early warning.

[0119] The method of combining the intrinsic correlation between multi-source information to effectively fuse the approximate subset of the multi-source information set as described in S02.4 specifically includes:

[0120] S02.4.1. First, the dynamic system of the tunnel structure needs to be modeled. Controlled input and external noise The influence can be expressed by the state space model as:

[0121]

[0122] in and is the state transfer matrix and control input matrix of the system, is Gaussian white noise, representing the random disturbance of the system.

[0123] Output It can be expressed as:

[0124]

[0125] in is the observation matrix, is the observation noise.

[0126] S02.4.2. Estimate the system state via a recursive prediction and update procedure suitable for a linear Gaussian noise environment:

[0127] S02.4.21. Initial Setup: Setting the Initial State Estimate and its error covariance matrix :

[0128]

[0129] S02.4.22. Prediction step: Estimation based on current state and input Predict the state at the next time point :

[0130]

[0131] And the prediction error covariance matrix :

[0132]

[0133] in is the process noise covariance.

[0134] S02.4.23. Update step: based on new observation data , update the state estimate:

[0135]

[0136] in is the gain matrix:

[0137]

[0138] And update the error covariance matrix:

[0139]

[0140] in is the observation noise covariance.

[0141] S02.4.24, Multi-source information fusion: Repeat the above update steps for all sensors 𝑖, and combine the observation results of each sensor, calculate the weighted average or other fusion method, and obtain the final fusion state estimate.

[0142] S02.4.3. Based on the inherent correlations between different sensor data, such as temperature changes that may affect the stress distribution of the tunnel structure, all relevant inherent correlations are integrated in the fusion algorithm:

[0143] S02.4.31. Incorporate the correlation between sensor data into the state space model to make the tunnel stress Affected by temperature The impact of expanding the tunnel state vector To include temperature information, or in the state transfer matrix The temperature-dependent terms are introduced in .

[0144] State vector expansion:

[0145]

[0146] in is the stress state of the tunnel, is the temperature state of the tunnel.

[0147] State transition matrix Including the relationship between temperature and stress:

[0148]

[0149] Represents the direct effect of temperature on stress.

[0150] S02.4.32. Mining the correlation between different sensor data through historical data or real-time data:

[0151] Represent multi-source data as a matrix 𝑋, where each row represents the observation value of a different sensor and each column represents the observation at a different time point. Calculate the covariance matrix of the matrix 𝑋:

[0152]

[0153] The covariance matrix is ​​subjected to eigenvalue decomposition to extract the main eigenvectors, which represent the most important direction of change in the data and are used to reveal the implicit correlation between the data of each sensor. The main eigenvectors are introduced into the state model as new variables.

[0154] S02.4.33. Divide the multi-source data into two groups, such as stress sensor data and temperature sensor data, and form two matrices respectively X 1 and X 2. Find the linear combination that maximizes the correlation between the two data sets by solving linear algebraic equations:

[0155]

[0156] The obtained related variables are introduced into the state model as new state variables.

[0157] S02.4.4. During the recursive processing, as new observations arrive, the noise covariance matrix needs to be dynamically adjusted and To adapt to the dynamic changes of the system.

[0158] S02.4.41. Process noise covariance matrix :

[0159] Initial setting through historical data analysis or experience , over time, based on the current state estimation error Make continuous recursive corrections:

[0160]

[0161] in α is the smoothing factor, usually between 0 and 1.

[0162] S02.4.42. Observation noise covariance matrix :

[0163] Usually the initial setting is based on the noise characteristics of the sensor , in the recursive process, the actual error observed by the sensor is dynamically adjusted :

[0164]

[0165] in β is the smoothing factor.

[0166] S02.4.43. Through the feedback mechanism in recursive calculation, adjust the parameters of the state model in S02.4.2 so that the model can better adapt to the changes in real-time data.

[0167] After each update, the state estimation error is calculated and the model parameters of the filter are modified:

[0168]

[0169] if If the value exceeds the preset threshold, it may indicate that some parameters of the model need to be adjusted, such as the observation matrix Or the state transition matrix .

[0170] Use feedback mechanism to dynamically adjust the filter gain :

[0171]

[0172] in γ It is the adjustment coefficient. According to the error feedback, the filter parameters are dynamically adjusted to gradually optimize the fusion effect.

[0173] The intelligent evaluation algorithm includes a neural network and a fuzzy logic algorithm, and the analysis tool includes an expert system and numerical analysis;

[0174] The health status assessment module specifically includes:

[0175] S03.1. Perform multi-level analysis on the high-quality large data set obtained after processing based on the multi-source data fusion module, adaptively process complex multi-source data based on big data and the neural network learning algorithm, learn complex structural state distribution characteristics and variable patterns from a large amount of historical data, and analyze the spatiotemporal evolution laws between data and physical observation phenomena;

[0176] S03.2. Combine multiple physical analytical models to establish the mapping relationship between the structural state response and the physical analytical model, and verify its effectiveness through numerical analysis. Through fine meshing and mechanical calculation of the tunnel structure, the stress and deformation of the tunnel structure under various load conditions are simulated to accurately predict the response and potential weak points of the structure;

[0177] S03.3. The numerical analysis results constitute a twin data set of the tunnel structure, and the parameters of the physical analytical model are dynamically corrected in combination with the data-driven method to accurately realize the structural state response, correct the mutual feedback of the physical analytical model to improve the data-driven quality, and establish a physical model-data-driven mutual feedback collaborative and dynamic progressive hierarchical analysis mechanism;

[0178] S03.4. Use fuzzy logic algorithm to process the uncertainty and fuzzy information in the output results after the hierarchical analysis mechanism is processed, and the information entropy of the structural state space perception is reduced to the greatest extent. By setting the fuzzy set and rules of the tunnel structure health status, the health level and risk level of the structure are evaluated to provide clear evaluation results;

[0179] S03.5. The expert system integrates the knowledge and experience of engineering experts, builds a rule base and a knowledge base, automatically diagnoses and evaluates the health status of the tunnel structure, identifies potential structural problems and failure modes, and systematically compares real-time data with historical data through historical data comparison and analysis, identifies long-term changes and periodic change trends of the tunnel structure and potential degradation problems, and evaluates the life and stability of the structure;

[0180] S03.6. Generate a detailed assessment report based on the above assessment results, including structural health score, potential risk locations, possible fault types, maintenance recommendations, etc. The report details the potential risk locations in the tunnel structure, identifies possible problem areas in the structure through analysis and evaluation of the monitoring data, and provides a detailed description of fault types, such as cracks, deformation, leakage, etc. in the tunnel, by analyzing the monitoring data and assessment results, and provides corresponding specific maintenance recommendations to help relevant personnel take maintenance and repair measures in a timely manner to ensure the safety and stability of the tunnel structure.

[0181] S03.7. The visualization display function uses a variety of charts to intuitively display the structural health score and the trend of changes in monitoring data. At the same time, it uses a 3D model to display the overall situation and evaluation results of the tunnel structure. It supports rotation, zooming and multi-angle viewing, and provides an intuitive display of the structural health status.

[0182] The multi-level analysis of the high-quality large data set obtained by the multi-source data fusion module in the health status assessment module described in S03.1 specifically includes:

[0183] S03.1.1. Before conducting the multi-level analysis, the data set needs to be processed first. After the data is processed by the multi-source fusion module, a high-quality data set is formed. The data needs to be further prepared and organized for more in-depth analysis.

[0184] S03.1.11. Data Standardization:

[0185] The data of different sensors Normalization and standardization are used to ensure that data of different dimensions are comparable. Normalization usually uses min-max scaling:

[0186]

[0187] S03.1.12 Data stratification:

[0188] The data is divided into three levels: short-term, medium-term and long-term according to the time span. The short-term data is used to capture instantaneous changes, and the long-term data is used to reflect trends. The data is stratified according to the spatial location of the sensor. For example, different parts of the tunnel can be divided into top, side wall and bottom data, and then analyzed separately.

[0189] S03.1.2. Multi-level analysis involves analyzing data at different levels, such as time level, spatial level, and feature level, to fully understand the status of the tunnel structure.

[0190] S03.1.21, Time level:

[0191] The sliding window technique is used to analyze the changing trend of the short-term data, the data is smoothed by calculating the moving average, and the difference method is used to detect the change point:

[0192]

[0193] If Δ x t If it exceeds a certain threshold, it is marked as a potential outlier.

[0194] The ARIMA (AutoRegressive Integrated Moving Average) model is used for long-term trend analysis of time series. The ARIMA model can capture the trend and seasonal components of the data.

[0195]

[0196] Among them is The autoregressive coefficient, θ is the moving average coefficient, is the error term, and the best one is selected by AIC (AkaikeInformation Criterion) p , q parameter.

[0197] S03.1.22, spatial hierarchy:

[0198] By generating a spatial heat map of the sensor data to visualize the distribution of stress, temperature, humidity, etc. in each part of the tunnel structure, the heat map can reveal potential abnormal areas in the tunnel. By using spatial statistics such as Moran's I to detect the spatial autocorrelation of the data, it can be determined whether there is regional abnormal behavior:

[0199]

[0200] in is a weight matrix, representing the spatial adjacency relationship.

[0201] S03.1.23, Feature level:

[0202] Use PCA (Principal Component Analysis) or LDA (Linear Discriminant Analysis) to extract key features and reduce feature dimensions to reduce noise and improve the generalization ability of the model. Then combine wavelet transform to perform multi-scale analysis on the data and decompose different frequency components. The low-frequency components usually represent the trend of the data, and the high-frequency components usually represent instantaneous changes.

[0203] The mapping relationship between the structural state response and the physical analytical model established by means of numerical analysis in the health status assessment module described in S03.2 specifically includes:

[0204] S03.2.1. Geometric modeling of the tunnel structure. Abaqus modeling software is used to create a three-dimensional geometric model of the tunnel. The three-dimensional geometric model of the tunnel accurately reflects the actual structural characteristics of the tunnel, such as cross-sectional shape, length, lining thickness, etc. The geometric characteristics - tunnel lining: define the thickness and shape of the tunnel lining, consider key positions such as the arch crown and arch foot, and the stratum - consider the stratum model around the tunnel, define the layers, thickness and physical properties of the stratum.

[0205] Material property definitions - Modulus of elasticity (E): The modulus of elasticity of concrete is typically in the range of 20-40 GPa; Poisson's ratio (ν): Typically 0.2-0.3; Density (ρ): For concrete, a typical value is 2400 kg / m³.

[0206] Boundary conditions and load conditions - Boundary conditions: fix the two ends of the tunnel or specify the contact conditions between the foundation and the tunnel. Set typical boundary conditions such as fixed bottom and free sliding top; Load conditions: consider factors such as deadweight, formation pressure, static load and dynamic load. The load can be a uniform load or a concentrated load, depending on the actual working conditions.

[0207] S03.2.2. Perform finite element meshing on the three-dimensional geometric model of the tunnel and select the meshing type unit: tetrahedron or hexahedron units can be used for three-dimensional structural models. Tetrahedron units are suitable for complex geometric shapes and are used to set lining joints, while hexahedron units are more efficient in simple geometries and are used to set tunnel sections. For thin shell structures, such as tunnel linings, shell units are used for meshing to improve calculation accuracy.

[0208] Mesh encryption strategy - local encryption: partially encrypt the mesh in stress concentration areas such as the vault and arch foot. The typical unit size is in the range of 5-10 cm, while other areas can use a coarser mesh of 20-50 cm; overall mesh division: ensure mesh quality, avoid excessive distortion, and ensure regular unit shape.

[0209] S03.2.3. Numerical simulation and analysis.

[0210] S03.2.31. Static analysis: Use the linear finite element method to solve the stress and deformation of the tunnel under static load, output the stress field and displacement field, and identify the stress concentration area;

[0211]

[0212] in, K is the structural stiffness matrix, i is the displacement vector, F is the load vector.

[0213] S03.2.32. Nonlinear analysis includes models using nonlinear materials, such as the plastic damage model for concrete, simulating the behavior of the structure under extreme loads, accounting for geometric nonlinearity, and considering large deformation effects, using an updated stiffness matrix for iterative solution;

[0214] S03.2.33. Dynamic analysis: Calculate the natural frequencies and mode shapes of the tunnel structure to determine the vibration characteristics of the structure;

[0215]

[0216] in, w is the natural frequency, M is the mass matrix, is the mode shape vector.

[0217] S03.2.34 Transient analysis: Simulates the response of a tunnel to dynamic loads such as seismic waves, using explicit or implicit time integration methods such as the Newmark-beta method.

[0218] S03.2.4. Verification of numerical simulation results:

[0219] S03.2.41. Use field monitoring data such as measured values ​​of stress gauges and accelerometers to compare with numerical simulation results, calculate the error between simulation results and measured data, conduct error analysis, and analyze the sources of error, such as model simplification, boundary condition assumptions, etc.;

[0220] S03.2.42. Perform mesh convergence test, repeat simulation at different mesh densities, ensure that the results gradually converge with mesh refinement, record stress and displacement of key points such as stress concentration points, draw convergence curves, and ensure that the simulation results tend to be stable after a certain mesh density;

[0221] S03.2.43. Perform parameter sensitivity tests, change material parameters such as elastic modulus, Poisson's ratio, and boundary conditions one by one, observe the changes in simulation results, determine the key parameters of the model, and adjust these parameters to better match the experimental data.

[0222] S03.2.5, Model optimization and iteration: Based on the error analysis results of step S03.2.41, gradually adjust the material parameters, boundary conditions or load assumptions to reduce the error between the numerical simulation and the experimental results. After each model adjustment, repeat the simulation and experimental comparison until the error between the simulation results and the experimental data is within an acceptable range, usually less than 5%-10%. As more field data is collected, the model parameters are regularly updated and retrained to make the model more adaptable.

[0223] The early warning module integrates the multi-source data and combines the multi-level early warning algorithm to conduct real-time monitoring and risk assessment of the tunnel structure to ensure that early warning signals are issued in time before potential risks occur, specifically including:

[0224] S04.1. Use threshold algorithms to set safety thresholds for tunnel structure health status. For example, for key parameters such as stress and deformation, when the monitoring data exceeds or approaches the preset threshold, an early warning signal is immediately triggered. This is suitable for scenarios with clear risk indicators to ensure real-time monitoring of the key status of the tunnel structure.

[0225] S04.2. The probabilistic risk intelligent assessment algorithm based on Bayesian network integrates the uncertainty factors in the multi-source information set, such as material aging, environmental changes, construction quality, etc., sets the fuzzy set and rules of the tunnel structure health status to build a risk assessment model for the tunnel structure, updates the structural health status data in real time, dynamically calculates the risk probability distribution of the structure, and generates warning signals of different levels, such as low risk, medium risk and high risk. Each level corresponds to different response measures to ensure accurate warning in a complex risk environment;

[0226] S04.3. Combined with the complete set of tunnel structure status information output by the multi-source data fusion module, the deep learning algorithm is used to perform anomaly detection again, and the abnormal patterns in historical data and real-time data are identified by training the model, so as to automatically detect potential risks that may not be identified by traditional algorithms and early warning misjudgments caused by data accumulation errors of structural state information after processing by multiple modules, thereby improving the perception ability of the early warning system for complex and nonlinear risks;

[0227] S04.4. The multi-channel notification system uses IoT technology to connect smart devices such as smart watches, mobile applications, alarms, etc. to transmit warning signals in real time, and designs redundant notification mechanisms to ensure multi-channel coverage. Based on the big data integration platform, the notification system can dynamically evaluate the priority of warning information to ensure that relevant personnel can receive the dynamic response strategy recommendations in a timely manner and process the warning information, thereby improving the timeliness and reliability of notifications;

[0228] S04.5. Keep detailed records of each warning event, including warning time, trigger parameters, over-limit range, response measures and other information, to provide a basis for subsequent analysis, fault tracing and maintenance. Review and analyze the events, evaluate the rationality of threshold settings and the effectiveness of response measures, and feed back the analysis results to the threshold setting module to optimize subsequent warning effects.

[0229] The multi-level warning algorithm in the warning module described in S04.1 includes parameter selection and threshold setting, real-time monitoring and threshold judgment. The specific steps are as follows:

[0230] S04.1.1. The early warning module relies on the multi-source data fusion module to obtain a multi-source information set including the external environment, structural performance status, material characteristics, etc., and input multi-source data. Through these data, the system identifies parameters that are critical to the health status of the tunnel structure, such as stress, strain, crack width, temperature and humidity, displacement, etc.;

[0231] S04.1.2. Calculate the limit values ​​of the tunnel structure under different working conditions by combining the tunnel design specifications, construction standards, and material mechanics and structural mechanics theories. Identify the normal parameter ranges and limit values ​​of the tunnel under different environmental conditions by relying on the historical data comparison and analysis function. Simulate the stress and deformation of the tunnel structure under different loads and environmental conditions by using the numerical analysis and simulation functions to set a reasonable safety threshold.

[0232] S04.1.3. The system compares the real-time collected monitoring data with the pre-set safety threshold. When the real-time value of a key parameter approaches or exceeds the set threshold, the system determines the risk level according to the degree of excess and triggers the corresponding warning signal;

[0233] S04.1.4. Based on the adaptive analysis function of the health status assessment module and the adaptive assessment system data feedback of the early warning system, the system dynamically adjusts the safety thresholds of the key parameters, learns new risk patterns and automatically optimizes the threshold settings to improve the accuracy and timeliness of the early warning.

[0234] The response strategy recommendation module generates specific maintenance and emergency response recommendations based on the real-time analysis results of the fused data. By integrating the historical records and maintenance experience of multi-source data, it provides detailed inspection items, repair methods, and the required tools and materials. At the same time, it provides emergency response plans for high-risk warnings, such as personnel evacuation, traffic control, and emergency reinforcement measures to ensure rapid and effective response in critical situations. The response strategy recommendation module has an adaptive optimization function, which forms a closed-loop management through cyclic learning and data feedback, and continuously adjusts and optimizes the response strategy to adapt to the ever-changing risk situation.

[0235] The beneficial effects of the present invention are as follows: the present invention provides a tunnel structure health monitoring and early warning system with multi-source twin data fusion. By acquiring and fusing the equipment status data, tunnel environment data, traffic condition data and various monitoring data of the tunnel structure in real time, the internal and external data of the tunnel can be comprehensively and specifically acquired, thereby improving the accuracy of tunnel health status monitoring. The health status assessment module can quickly assess the health status of the tunnel structure based on the fused tunnel monitoring data, identify potential structural problems of the tunnel, generate a detailed assessment report based on the assessment results, and visualize the tunnel structure. The staff can quickly understand the health status of the tunnel structure through the assessment report, thereby improving the comprehensiveness and applicability of tunnel structure health status monitoring. The early warning module forms an early warning result based on the assessment result, and notifies the staff in time when a tunnel structure health problem is monitored. The early warning module can also provide a detailed description of the fault type, such as cracks, deformation, leakage, etc. of the tunnel, and provide corresponding specific maintenance suggestions, thereby improving the decision-making efficiency of the command department, and helping relevant personnel to take maintenance and repair measures in time to ensure the safety and stability of the tunnel structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0236] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0237] Figure 1 It is a system structure diagram of the tunnel structure health monitoring and early warning system with multi-source twin data fusion of the present invention.

[0238] Figure 2 It is a schematic diagram of the data acquisition module.

[0239] Figure 3 It is a schematic diagram of the multi-source data fusion module.

[0240] Figure 4 It is a schematic diagram of the health status assessment module.

[0241] Figure 5 It is a schematic diagram of the early warning module. DETAILED DESCRIPTION

[0242] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings, but it should be understood that the present invention can be implemented in various forms and should not be limited by the exemplary embodiments described herein; on the contrary, this embodiment is provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to the application or technical personnel in the field. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0243] Example 1: Reference Figure 1

[0244] like Figure 1 As shown, the present invention discloses a tunnel structure health monitoring and early warning system with multi-source twin data fusion, including a data acquisition module, a multi-source data fusion module, a health status assessment module and an early warning module. The single vehicle model training includes:

[0245] The data acquisition module includes a sensing technology module and a data transmission unit. The sensing technology module is used to collect equipment status data, tunnel environment data, traffic status data and various monitoring data of the tunnel structure to monitor the tunnel, forming multi-source data of the tunnel. The various monitoring data include parameters such as stress, deformation, temperature, humidity, etc. The data acquisition module includes various sensing devices, such as strain gauges, accelerometers, temperature sensors, humidity sensors, displacement sensors, vibration sensors and environmental monitoring sensors, etc.; the data transmission unit is used to transmit the collected data to the data processing center in real time, and has a data preprocessing function. The data preprocessing function includes preliminary filtering and preliminary analysis of the data to ensure the accuracy and reliability of the data. The data processed by the data processing center is transmitted to the multi-source data fusion module.

[0246] The multi-source data fusion module includes a data integration platform module, a data fusion and processing algorithm, and data correction, which are used to integrate and process the collected multi-source tunnel data, thereby improving the accuracy and consistency of the data. The data integration platform is used to uniformly manage and coordinate data from different sensors and monitoring equipment; the data fusion and processing algorithm effectively processes and fuses the multi-source tunnel data through methods such as Kalman filtering, Bayesian reasoning, and principal component analysis, eliminates redundant information and data noise, and improves the reliability and effectiveness of the data. The data correction is used to deal with data missing problems caused by equipment failure or data loss to ensure data integrity.

[0247] The health status assessment module includes an intelligent assessment algorithm, a multi-level health assessment, a historical data comparison and analysis, a detailed assessment report generation and a visual display, and is used to evaluate the health status of the tunnel structure based on the fused data. On the basis of the fused data of the multi-source data fusion module, the health status assessment module uses an intelligent assessment algorithm to evaluate the health status of the tunnel structure and identify potential structural problems. The intelligent assessment algorithm and the multi-level health assessment include a variety of analysis tools and intelligent assessment algorithms, such as finite element analysis, neural network algorithm, fuzzy logic algorithm and expert system, to provide a multi-level health assessment. The historical data comparison and analysis discovers and evaluates the structural change trend by comparing historical data with real-time data; the detailed assessment report generation generates a detailed tunnel health assessment report after analysis, and the tunnel health assessment report includes a structural health score, potential risk locations and possible fault types, and provides corresponding maintenance suggestions for staff; the module also supports the visual display, and intuitively displays the health status and assessment results of the tunnel structure in the form of charts, 3D models, etc., so that relevant personnel can understand and deal with structural problems in a timely manner.

[0248] The early warning module includes a multi-level early warning algorithm, dynamic response strategy recommendations and a multi-channel notification system. According to the health status assessment results, the multi-level early warning algorithm is used to form an early warning result, and relevant personnel are notified through various channels so that corresponding maintenance measures can be taken in time. The multi-level early warning algorithm includes a threshold algorithm and a probabilistic risk intelligent assessment algorithm to identify different types and levels of structural risks. By setting thresholds, various parameters in the tunnel are monitored, and graded early warnings are issued when the thresholds are approached or exceeded. Early warning signals of different levels are generated according to the degree of risk to ensure that relevant personnel can take appropriate measures according to the risk level; the multi-channel notification system sends early warning information to relevant personnel through various channels such as text messages, e-mails, mobile applications and Internet of Things devices; the response strategy recommendation function provides specific maintenance and emergency response suggestions based on the different types of early warning signals to help relevant personnel quickly and effectively respond to potential structural problems and dynamically adjust the tunnel structure.

[0249] Example 2: Reference Figure 2

[0250] like Figure 2 As shown, the data collected by the sensing technology module of the data acquisition module include tunnel structure health monitoring data and geological radar data, wherein the tunnel structure health monitoring data includes tunnel structural parameters, equipment status, tunnel environment and stress and strain, and the geological radar data includes construction status, address information, traffic conditions and rock mass status. The data collected by the sensing technology module specifically includes:

[0251] S01.1. Monitor the stress and strain of the tunnel structure under different load conditions through stress and strain sensors to capture structural deformation information;

[0252] S01.2. Install strain gauges and displacement sensors to record the crack distribution, characteristics and expansion patterns of the tunnel structure, and provide early warning information on potential risks of the structure;

[0253] S01.3. Use temperature sensors to monitor temperature changes inside the tunnel and in the surrounding environment, and assess the impact of temperature on tunnel structural materials and overall stability;

[0254] S01.4. Monitor the humidity level inside and around the tunnel through humidity sensors to detect corrosion and other potential impacts that a humid environment may have on the tunnel structure;

[0255] S01.5. Monitor the displacement and settlement of the tunnel structure through displacement sensors to ensure the overall stability and long-term safety of the structure;

[0256] S01.6. Capture the vibration data of the tunnel structure through vibration sensors and analyze the impact of external environmental factors on the tunnel structure;

[0257] S01.7. The geological radar data acquisition unit uses geological radar technology to conduct non-destructive testing of the rock mass and geological environment around the tunnel, obtain key information such as geological layers, cracks and water flows, and ensure the coordination and safety of the tunnel and the surrounding geological environment.

[0258] The data transmission unit also includes:

[0259] S01.8. The data transmission unit transmits the collected data to the data processing center in real time, supporting wired transmission and wireless transmission;

[0260] S01.9. The data processing center includes preliminary data filtering and preliminary analysis, and performs data preprocessing on the collected real-time data.

[0261] The external environmental factors may include traffic loads, earthquakes, etc. By capturing the vibration data in the tunnel environment and the external environmental factors, data support can be provided for the formulation of vibration control and protective measures for the tunnel.

[0262] The data transmission unit ensures stable transmission and real-time accessibility of monitoring data in different environments by supporting wired transmission and wireless transmission, adapts to the changing environment of tunnel construction and operation, ensures efficient collection and transmission of data, and provides a reliable data source for subsequent data processing and fusion.

[0263] The preliminary data filtering includes outlier detection and noise filtering;

[0264] S01.9.1. Outlier Detection:

[0265] S01.9.11, Statistical-based methods: For each sensor data (such as stress, temperature), assume that its observations over a period of time follow a normal distribution. Detect outliers by calculating the z-score of each data point:

[0266]

[0267] in x i is the observed value at a certain moment, μ is the mean of historical data, σ is the standard deviation, if z If the value exceeds a certain threshold, the data point is considered to be an outlier.

[0268] S01.9.12, based on the sliding window method: using the sliding window to calculate the local mean and standard deviation , and detect whether the current data significantly deviates from the statistical characteristics within the window:

[0269]

[0270] Among them, N is the window size of the sliding window, t is the sliding window time point, is the observed value at that time point. If the current data x t If the deviation from the mean within the window exceeds a certain threshold, it is marked as an anomaly.

[0271] S01.9.2. The noise filtering includes:

[0272] S01.9.21, Low-pass filtering: By setting a low-pass filter, high-frequency noise in the sensor data is filtered out while retaining the effective components in the signal.

[0273] Let the sampling frequency bef s , the cutoff frequency of the selected filter f c and filter order n , the filter transfer function for:

[0274]

[0275] in w c =2Πf c .

[0276] By applying the discretized transfer function to the sensor data sequence, high frequency noise can be eliminated.

[0277] S01.9.22, Median filter: used to remove impulse noise.

[0278] Remove isolated noise points by replacing the central data point with the median value in the data window:

[0279]

[0280] 2 of them k +1 is for window size, is the central data point value after replacement;

[0281] S01.9.3, the data feature extraction includes:

[0282] S01.9.31, Time Domain Feature Extraction: Calculate Mean and standard deviation , and use skewness and kurtosis to capture the asymmetry and peak degree of the data distribution:

[0283]

[0284]

[0285] S01.9.32, Frequency domain feature extraction: Convert time domain data to frequency domain through Fast Fourier Transform FFT to analyze the frequency components of the signal:

[0286]

[0287] in is a discrete-time signal, is the time index, yes The Fourier transform of is the frequency index, is an imaginary unit, satisfyingj 2 =-1, N is the sequence Length;

[0288] The power spectral density (PSD) is calculated to identify the main frequency components and their corresponding amplitudes. The PSD can be obtained by calculating the autocorrelation function of the signal and performing a Fourier transform on it:

[0289]

[0290] Where R xx (m) is the autocorrelation function of the signal.

[0291] S01.9.4. The trend analysis includes:

[0292] S01.9.41. Linear Regression: Analyze time series trends in sensor data using linear regression , estimate the slope and intercept of the trend to determine whether there is a long-term change in the data:

[0293]

[0294] in is the intercept, is the slope, is the error term, and the parameter is estimated by the least squares method. and .

[0295] S01.9.42 Moving Average: Calculate the moving average of data to smooth short-term fluctuations and show long-term trends.

[0296]

[0297] in, is the moving average at time t, is the window size of the moving average, is the observation at time i, for a given window size M , the moving average helps identify potential trend changes and provides a basis for further analysis.

[0298] Example 3: Reference Figure 3

[0299] like Figure 3 As shown, the data integration platform includes multi-source data fusion, data processing algorithms, data correction and completion, and unified management and coordination of data. The specific steps include:

[0300] S02.1. The data integration platform module detects a subset of a multi-source information set according to various sensing technologies in the data acquisition module, wherein the multi-source information set includes a real information set of an external environment, a real information set of a structural performance state, and a real information set of a material characteristic state;

[0301] S02.2, reducing the dimension of the multi-source information set by dimensionality reduction technology, eliminating redundant information, simplifying the data set structure, and extracting main features, while applying sufficient constraints based on known tunnel physical information, and outputting an approximate subset of the multi-source information set;

[0302] S02.3. hierarchically process the information of each scale in the approximate subset of the multi-source information set by a multi-scale analysis method, perform precision correction on the information of each scale by combining the spatial resolution, temporal resolution and measurement accuracy of the sensor, and fuse the calibration information at each scale based on the multi-scale reconstruction technology;

[0303] S02.4, estimate the dynamic system in a noisy environment through a recursive algorithm, combine the inherent correlation between the multi-source information, effectively fuse the approximate subset of the multi-source information set, infer the structural state under certain constraints, and dynamically compare it with the real information set detected by the sensing technology module, combine prior knowledge and real-time data, make probability estimation and decision, recursively modify the constraints, and construct complete tunnel structure state information;

[0304] S02.5. The data correction based on the big data integration platform is used to monitor the quality of multi-source data, evaluate and mark the data, detect and correct errors and deviations that may occur in the data collection process of sensors and monitoring equipment, automatically identify abnormal data and make corrections, and handle data missing problems caused by equipment failure or data loss by inferring and completing missing data based on historical big data and maintenance records.

[0305] The data integration platform module also includes unified management and coordination, real-time data collection and efficient data storage management;

[0306] Effectively collect and integrate information from various data sources through standardized interfaces and protocols;

[0307] Support real-time data collection, instantly obtain data from various sensors and monitoring equipment, and perform preliminary processing and storage;

[0308] Distributed storage technology is used to ensure efficient storage and fast access to data, and to support the storage and retrieval of historical data.

[0309] The data integration platform is characterized by unified management and coordination, real-time data acquisition and efficient data storage management. Through standardized interfaces and protocols, the platform can effectively collect and integrate information from various data sources to ensure seamless connection and unified management of data. The data acquisition platform supports real-time data acquisition and can instantly obtain data from various sensors and monitoring equipment, and perform preliminary processing and storage, ensuring the real-time and accuracy of monitoring data. By adopting distributed storage technology, efficient storage and rapid access to data are ensured, providing powerful data management functions, capable of processing massive data, and supporting the storage and retrieval of historical data.

[0310] In a multi-source information collection, different types of data may be collected at different time and space scales. For example, a temperature and humidity sensor may collect data every hour, while a displacement sensor may collect data every minute. Therefore, it is necessary to perform time alignment and spatial interpolation on the data to ensure that different data sources are unified at the same time and space scales.

[0311] Time alignment: Resample data of different frequencies to the same time node through methods such as linear interpolation or spline interpolation.

[0312] Spatial alignment: If the data sources are located in different locations, such as temperature and humidity sensors and surface deformation measurement equipment, the data can be mapped to the same spatial reference frame through spatial interpolation based on the spatial location and influence range of the sensor layout.

[0313] Dimensionality reduction technology is used to reduce the dimensionality of multi-source information sets to reduce dimensions, eliminate redundant information, and simplify data structures. The core idea is to project high-dimensional data into a low-dimensional subspace while retaining the direction with the largest variance in the data, thereby extracting the most important features.

[0314] S02.2.1. Assume that the multi-source information set is represented as a matrix ,in is the sample size, is the number of features. Each feature is centered and standardized, and the mean is calculated and standard deviation , convert the original data into a standardized matrix :

[0315]

[0316] Calculate the covariance matrix :

[0317]

[0318] Covariance matrix Perform eigenvalue decomposition and obtain the eigenvalue and the corresponding eigenvector The eigenvalue indicates the variance of the principal component, and the eigenvector indicates the direction of the principal component. Sort by eigenvalue from large to small, k The largest eigenvalues ​​and their corresponding eigenvectors form a new low-dimensional subspace.

[0319] The standardized data matrix Projected into this low-dimensional subspace. Assuming that the previous k feature vector , then the new data is expressed as:

[0320]

[0321] in, is a The matrix contains the data after dimensionality reduction. This matrix retains the main features with the largest variance in the original data, thus simplifying the data set while retaining the most important information.

[0322] In the process of dimensionality reduction, physical constraints need to be imposed in combination with known physical information of the tunnel (such as geometric parameters, material properties, etc.). The purpose of combining physical constraints is to ensure that the results after dimensionality reduction are not only statistically valid, but also consistent with the structural characteristics and behavior of the tunnel in the actual physical sense.

[0323] S02.2.2. Combined tunnel geometry parameters

[0324] The geometry and dimensions of the tunnel (such as diameter, length, curvature, etc.) are known and can serve as important constraints in the dimensionality reduction process. For example:

[0325] Shape constraint: The cross section of a tunnel is usually circular or elliptical, which can be used as a geometric constraint. During the dimensionality reduction process, it can be ensured that the projected data still conforms to this basic shape. For example, if the distribution of a cross-section point cloud data deviates from a circle after dimensionality reduction, it is necessary to adjust it back to the actual tunnel shape through geometric correction.

[0326] Position constraints: The position of the tunnel in space can also be used as a constraint. For example, the axis of the tunnel is usually continuous and smooth, which should be maintained in the reduced-dimensional data. The PCA-projected data can be adjusted to always be distributed along this axis, thus conforming to the overall geometric characteristics of the tunnel.

[0327] S02.2.3. Combining material properties

[0328] Tunnel material properties (such as concrete strength, modulus, stress-strain curve, etc.) can also be used as constraints:

[0329] Material strength constraints: When evaluating stress and deformation data, if the reduced-dimensional data show that the stress exceeds the yield strength of the material or the deformation exceeds the allowable strain range of the material, these data points can be marked as abnormal or corrected to ensure that the reduced-dimensionality results are within a reasonable material range.

[0330] Material aging and erosion constraints: For long-term tunnels, the materials may change due to aging or erosion from the external environment. These physical properties can be used as constraints in the dimensionality reduction process. For example, if environmental data (such as temperature and humidity) show that erosion or aging is high in certain areas, the reduced data should be able to reflect the performance changes in these areas.

[0331] S02.2.4. Combining structural mechanics constraints

[0332] Tunnel structures exhibit specific mechanical properties during stress and deformation, such as the correlation between stress distribution and displacement. These properties can be obtained through numerical simulation, structural analysis and field observation, and used as constraints in dimensionality reduction:

[0333] Stress-strain relationship constraint: There is a clear physical relationship between the stress and deformation of the tunnel structure. According to the mechanical model of the material (such as elastic model, elastoplastic model, etc.), during the dimensionality reduction process, it can be ensured that the reduced data maintains the relationship between stress and deformation. For example, the principal components after PCA processing should be able to reflect the change pattern of stress and deformation at various locations in the tunnel that conforms to the material mechanics theory.

[0334] Deformation mode constraints: Tunnels may be affected by various factors during operation, such as surface settlement, groundwater changes, and vehicle loads. Based on these known factors and the force pattern of the structure, the data can be corrected during the dimensionality reduction process to conform to typical tunnel deformation patterns (such as longitudinal settlement, lateral contraction, etc.). For example, if the data after dimensionality reduction shows deformation that does not conform to physical laws (such as unreasonable and drastic changes), it can be corrected based on the mechanical model of the tunnel.

[0335] S02.2.5. Combining historical data and prior knowledge

[0336] Historical monitoring data and prior knowledge from tunnel operations are also important sources of physical constraints:

[0337] Historical trend constraints: Tunnels accumulate a large amount of historical data during long-term monitoring, which records the deformation, stress changes and other characteristics of the tunnel. By comparing and analyzing these historical data with current data, it is possible to ensure that the dimensionality reduction results are consistent with the long-term trend during the dimensionality reduction process. For example, if historical data shows that a certain area has a slow settlement trend in the past few years, the dimensionality reduction data should be consistent with this trend.

[0338] Prior knowledge constraints: Based on engineering experience and the structural safety specifications for tunnel design, the range of variation of certain data is known. During the dimensionality reduction process, a constraint on the range of data variation can be set so that any principal component data that exceeds this range is marked or corrected.

[0339] These constraints make the dimensionality reduction results more physically meaningful, thereby generating more accurate approximate subsets of multi-source information sets.

[0340] Multiscale analysis methods are used to perform hierarchical processing on approximate subsets of multi-source information sets to capture data features at different scales and fuse information at each scale through multiscale reconstruction techniques.

[0341] S02.3.1. Combination of sensor resolution

[0342] Sensors used in tunnel monitoring usually have different spatial and temporal resolutions. Multiscale analysis methods can process the data in layers according to the resolution of the sensors:

[0343] Spatial resolution: For example, for sensors measuring tunnel lining stress, different spatial resolutions will affect the accuracy of the data. Through multi-scale analysis, data at different resolutions can be hierarchically integrated to unify the scale and improve accuracy.

[0344] Temporal resolution: Different types of sensors (such as accelerometers and displacement sensors) may collect data at different time intervals. Through multi-scale analysis, data at different time scales can be fused to ensure that the structural state is accurately reflected even on a short time scale.

[0345] S02.3.2. Accuracy Calibration

[0346] In the multi-scale analysis process, the accuracy of information at different scales is corrected by combining the measurement accuracy of the sensor and known physical parameters. Specifically, it includes:

[0347] Calibrate small scale based on large scale: By calibrating high-frequency (small scale) data on low-frequency (large scale), it is possible to ensure that local information is consistent with the global trend. For example, in tunnel deformation monitoring, the global deformation trend can correct the local stress fluctuation data.

[0348] Noise filtering and anomaly detection: Multiscale analysis can also be used to detect and filter data noise at different scales. For example, for vibration data of tunnel linings, wavelet transform can separate high-frequency signals caused by environmental noise or measurement errors and filter them to retain the actual structural response signal.

[0349] S02.3.3. Based on the layered data obtained from multi-scale analysis, the calibration information at each scale can be fused through multi-scale reconstruction technology to restore the overall structural state.

[0350] Multi-scale fusion: By reconstructing low-frequency and high-frequency components at different scales, the complete tunnel structure state containing information at all levels can be obtained. For example, by combining high-frequency components (local fluctuations) with low-frequency components (overall trends), comprehensive state information that conforms to the overall mechanical laws and reflects local anomalies can be reconstructed.

[0351] Fusion formula: Assumptions The weights at each scale are , then the multi-scale reconstruction can be expressed as:

[0352]

[0353] in, In the reconstruction process, the weight and accuracy of each sensor can be combined for weighted fusion to ensure that the final reconstruction result reflects the actual state of the tunnel to the greatest extent.

[0354] By applying multi-scale analysis and reconstruction techniques, physically meaningful structural features and status information can be extracted from complex multi-source information sets, thus providing accurate data support for tunnel health assessment and early warning.

[0355] The method of combining the intrinsic correlation between multi-source information to effectively fuse the approximate subset of the multi-source information set as described in S02.4 specifically includes:

[0356] S02.4.1. First, the dynamic system of the tunnel structure needs to be modeled. Controlled input and external noise The influence can be expressed by the state space model as:

[0357]

[0358] in and is the state transfer matrix and control input matrix of the system, is Gaussian white noise, representing the random disturbance of the system.

[0359] Output It can be expressed as:

[0360]

[0361] in is the observation matrix, is the observation noise.

[0362] S02.4.2. Estimate the system state via a recursive prediction and update procedure suitable for a linear Gaussian noise environment:

[0363] S02.4.21. Initial Setup: Setting the Initial State Estimate and its error covariance matrix :

[0364]

[0365] S02.4.22. Prediction step: Estimation based on current state and input Predict the state at the next time point :

[0366]

[0367] And the prediction error covariance matrix :

[0368]

[0369] in is the process noise covariance.

[0370] S02.4.23. Update step: based on new observation data , update the state estimate:

[0371]

[0372] in is the gain matrix:

[0373]

[0374] And update the error covariance matrix:

[0375]

[0376] in is the observation noise covariance.

[0377] S02.4.24, Multi-source information fusion: Repeat the above update steps for all sensors 𝑖, and combine the observation results of each sensor, calculate the weighted average or other fusion method, and obtain the final fusion state estimate.

[0378] S02.4.3. Based on the inherent correlations between different sensor data, such as temperature changes that may affect the stress distribution of the tunnel structure, all relevant inherent correlations are integrated in the fusion algorithm:

[0379] S02.4.31. Incorporate the correlation between sensor data into the state space model to make the tunnel stress Affected by temperature The impact of expanding the tunnel state vector To include temperature information, or in the state transfer matrix The temperature-dependent terms are introduced in .

[0380] State vector expansion:

[0381]

[0382] in is the stress state of the tunnel, is the temperature state of the tunnel.

[0383] State transition matrix Including the relationship between temperature and stress:

[0384]

[0385] Represents the direct effect of temperature on stress.

[0386] S02.4.32. Mining the correlation between different sensor data through historical data or real-time data:

[0387] Representing multi-source data as a matrix X , where each row represents the observation value of a different sensor and each column represents the observation at a different time point. X Calculate the covariance matrix:

[0388]

[0389] The covariance matrix is ​​subjected to eigenvalue decomposition to extract the main eigenvectors, which represent the most important direction of change in the data and are used to reveal the implicit correlation between the data of each sensor. The main eigenvectors are introduced into the state model as new variables.

[0390] S02.4.33. Divide the multi-source data into two groups, such as stress sensor data and temperature sensor data, and form two matrices respectively X 1 and X 2. Find the linear combination that maximizes the correlation between the two data sets by solving linear algebraic equations:

[0391]

[0392] The obtained related variables are introduced into the state model as new state variables.

[0393] S02.4.4. During the recursive processing, as new observations arrive, the noise covariance matrix needs to be dynamically adjusted and To adapt to the dynamic changes of the system.

[0394] S02.4.41. Process noise covariance matrix :

[0395] Initial setting through historical data analysis or experience , over time, based on the current state estimation error Make continuous recursive corrections:

[0396]

[0397] in α is the smoothing factor, usually between 0 and 1.

[0398] S02.4.42. Observation noise covariance matrix :

[0399] Usually the initial setting is based on the noise characteristics of the sensor , in the recursive process, the actual error observed by the sensor is dynamically adjusted :

[0400]

[0401] in β is the smoothing factor.

[0402] S02.4.43. Through the feedback mechanism in recursive calculation, adjust the parameters of the state model in S02.4.2 so that the model can better adapt to the changes in real-time data.

[0403] After each update, the state estimation error is calculated and the model parameters of the filter are modified:

[0404]

[0405] if If the value exceeds the preset threshold, it may indicate that some parameters of the model need to be adjusted, such as the observation matrix Or the state transition matrix .

[0406] Use feedback mechanism to dynamically adjust the filter gain :

[0407]

[0408] in γIt is the adjustment coefficient. According to the error feedback, the filter parameters are dynamically adjusted to gradually optimize the fusion effect.

[0409] Example 4: Reference Figure 4

[0410] like Figure 4 As shown, the health status assessment module specifically includes:

[0411] S03.1. Perform multi-level analysis and processing on the data set processed by the multi-source data fusion module through a neural network algorithm, learn the state distribution characteristics and variable patterns of complex tunnel structures from historical data, and analyze the spatiotemporal evolution laws between data and physical observation phenomena;

[0412] S03.2. Establish a mapping relationship between the structural state response and the physical analytical model by combining multiple physical analytical models, and verify its effectiveness through the numerical analysis method. By performing fine meshing and mechanical calculation on the tunnel structure, the stress and deformation of the tunnel structure under various load conditions are simulated to accurately predict the response and potential weak points of the structure;

[0413] S03.3. The numerical analysis results constitute a twin data set of the tunnel structure, and the parameters of the physical analytical model are dynamically corrected in combination with the data-driven method to accurately realize the structural state response, correct the mutual feedback of the physical analytical model to improve the data-driven quality, and establish a physical model-data-driven mutual feedback collaborative and dynamic progressive hierarchical analysis mechanism;

[0414] S03.4. Using the fuzzy logic algorithm to process the uncertainty and fuzzy information in the output results after the hierarchical analysis mechanism, the information entropy reduction of the structural state space perception is reduced to the greatest extent, and by setting the fuzzy set and rules of the tunnel structure health state, the health level and risk level of the structure are evaluated to provide clear evaluation results;

[0415] S03.5. The expert system integrates the knowledge and experience of engineering experts, builds a rule base and a knowledge base, automatically diagnoses and evaluates the health status of the tunnel structure, identifies potential structural problems and failure modes, and systematically compares real-time data with historical data through historical data comparison and analysis, identifies long-term changes and periodic change trends of the tunnel structure and potential degradation problems, and evaluates the life and stability of the structure;

[0416] S03.6. Generate a detailed assessment report based on the above assessment results, including structural health score, potential risk location, possible fault type and specific maintenance recommendations, and provide a detailed description of the fault type and corresponding maintenance recommendations by analyzing the monitoring data and assessment results;

[0417] S03.7. The visualization display function uses a variety of charts to intuitively display the structural health score and the trend of changes in monitoring data. At the same time, it uses a 3D model to display the overall situation and evaluation results of the tunnel structure. It supports rotation, zooming and multi-angle viewing, and provides an intuitive display of the structural health status.

[0418] The multi-level analysis of the high-quality large data set obtained by the multi-source data fusion module in the health status assessment module described in S03.1 specifically includes:

[0419] S03.1.1. Before conducting the multi-level analysis, the data set needs to be processed first. After the data is processed by the multi-source fusion module, a high-quality data set is formed. The data needs to be further prepared and organized for more in-depth analysis.

[0420] S03.1.11. Data Standardization:

[0421] The data of different sensors Normalization and standardization are used to ensure that data of different dimensions are comparable. Normalization usually uses min-max scaling:

[0422]

[0423] S03.1.12 Data stratification:

[0424] The data is divided into three levels: short-term, medium-term and long-term according to the time span. The short-term data is used to capture instantaneous changes, and the long-term data is used to reflect trends. The data is stratified according to the spatial location of the sensor. For example, different parts of the tunnel can be divided into top, side wall and bottom data, and then analyzed separately.

[0425] S03.1.2. Multi-level analysis involves analyzing data at different levels, such as time level, spatial level, and feature level, to fully understand the status of the tunnel structure.

[0426] S03.1.21, Time level:

[0427] The sliding window technique is used to analyze the changing trend of the short-term data, the data is smoothed by calculating the moving average, and the difference method is used to detect the change point:

[0428]

[0429] If Δ x t If it exceeds a certain threshold, it is marked as a potential outlier.

[0430] The ARIMA (AutoRegressive Integrated Moving Average) model is used for long-term trend analysis of time series. The ARIMA model can capture the trend and seasonal components of the data.

[0431]

[0432] in is the autoregressive coefficient, θ is the moving average coefficient, is the error term, and the best one is selected by AIC (AkaikeInformation Criterion) p , q parameter.

[0433] S03.1.22, spatial hierarchy:

[0434] By generating a spatial heat map of the sensor data to visualize the distribution of stress, temperature, humidity, etc. in each part of the tunnel structure, the heat map can reveal potential abnormal areas in the tunnel. By using spatial statistics such as Moran's I to detect the spatial autocorrelation of the data, it can be determined whether there is regional abnormal behavior:

[0435]

[0436] in is a weight matrix, representing the spatial adjacency relationship.

[0437] S03.1.23, Feature level:

[0438] Use PCA (Principal Component Analysis) or LDA (Linear Discriminant Analysis) to extract key features and reduce feature dimensions to reduce noise and improve the generalization ability of the model. Then combine wavelet transform to perform multi-scale analysis on the data and decompose different frequency components. The low-frequency components usually represent the trend of the data, and the high-frequency components usually represent instantaneous changes.

[0439] The mapping relationship between the structural state response and the physical analytical model established by means of numerical analysis in the health status assessment module described in S03.2 specifically includes:

[0440] S03.2.1. Geometric modeling of the tunnel structure. Abaqus modeling software is used to create a three-dimensional geometric model of the tunnel. The three-dimensional geometric model of the tunnel accurately reflects the actual structural characteristics of the tunnel, such as cross-sectional shape, length, lining thickness, etc. The geometric characteristics - tunnel lining: define the thickness and shape of the tunnel lining, consider key positions such as the arch crown and arch foot, and the stratum - consider the stratum model around the tunnel, define the layers, thickness and physical properties of the stratum.

[0441] Material property definitions - Modulus of elasticity (E): The modulus of elasticity of concrete is typically in the range of 20-40 GPa; Poisson's ratio (ν): Typically 0.2-0.3; Density (ρ): For concrete, a typical value is 2400 kg / m³.

[0442] Boundary conditions and load conditions - Boundary conditions: fix the two ends of the tunnel or specify the contact conditions between the foundation and the tunnel. Set typical boundary conditions such as fixed bottom and free sliding top; Load conditions: consider factors such as deadweight, formation pressure, static load and dynamic load. The load can be a uniform load or a concentrated load, depending on the actual working conditions.

[0443] S03.2.2. Perform finite element meshing on the three-dimensional geometric model of the tunnel and select the meshing type unit: tetrahedron or hexahedron units can be used for three-dimensional structural models. Tetrahedron units are suitable for complex geometric shapes and are used to set lining joints, while hexahedron units are more efficient in simple geometries and are used to set tunnel sections. For thin shell structures, such as tunnel linings, shell units are used for meshing to improve calculation accuracy.

[0444] Mesh encryption strategy - local encryption: partially encrypt the mesh in stress concentration areas such as the vault and arch foot. The typical unit size is in the range of 5-10 cm, while other areas can use a coarser mesh of 20-50 cm; overall mesh division: ensure mesh quality, avoid excessive distortion, and ensure regular unit shape.

[0445] S03.2.3. Numerical simulation and analysis.

[0446] S03.2.31. Static analysis: Use the linear finite element method to solve the stress and deformation of the tunnel under static load, output the stress field and displacement field, and identify the stress concentration area;

[0447]

[0448] in, K is the structural stiffness matrix, u is the displacement vector, F is the load vector.

[0449] S03.2.32. Nonlinear analysis includes models using nonlinear materials, such as the plastic damage model for concrete, simulating the behavior of the structure under extreme loads, accounting for geometric nonlinearity, and considering large deformation effects, using an updated stiffness matrix for iterative solution;

[0450] S03.2.33. Dynamic analysis: Calculate the natural frequencies and mode shapes of the tunnel structure to determine the vibration characteristics of the structure;

[0451]

[0452] in, w is the natural frequency, M is the mass matrix, is the mode shape vector.

[0453] S03.2.34 Transient analysis: Simulates the response of a tunnel to dynamic loads such as seismic waves, using explicit or implicit time integration methods such as the Newmark-beta method.

[0454] S03.2.4. Verification of numerical simulation results:

[0455] S03.2.41. Use field monitoring data such as measured values ​​of stress gauges and accelerometers to compare with numerical simulation results, calculate the error between simulation results and measured data, conduct error analysis, and analyze the sources of error, such as model simplification, boundary condition assumptions, etc.;

[0456] S03.2.42. Perform mesh convergence test, repeat simulation at different mesh densities, ensure that the results gradually converge with mesh refinement, record stress and displacement of key points such as stress concentration points, draw convergence curves, and ensure that the simulation results tend to be stable after a certain mesh density;

[0457] S03.2.43. Perform parameter sensitivity tests, change material parameters such as elastic modulus, Poisson's ratio, and boundary conditions one by one, observe the changes in simulation results, determine the key parameters of the model, and adjust these parameters to better match the experimental data.

[0458] S03.2.5, Model optimization and iteration: Based on the error analysis results of step S03.2.41, gradually adjust the material parameters, boundary conditions or load assumptions to reduce the error between the numerical simulation and the experimental results. After each model adjustment, repeat the simulation and experimental comparison until the error between the simulation results and the experimental data is within an acceptable range, usually less than 5%-10%. As more field data is collected, the model parameters are regularly updated and retrained to make the model more adaptable.

[0459] Example 5: Reference Figure 5

[0460] like Figure 5 As shown, the early warning module generates an early warning signal through the threshold algorithm and the probability risk algorithm in the early warning algorithm, notifies the staff through the multi-channel notification system, and provides specific suggestions and protective measures for the staff through the response strategy suggestion function. The early warning module specifically includes:

[0461] The early warning module integrates the multi-source data and combines the multi-level early warning algorithm to conduct real-time monitoring and risk assessment of the tunnel structure to ensure that early warning signals are issued in time before potential risks occur, specifically including:

[0462] S04.1. Use threshold algorithms to set safety thresholds for tunnel structure health status. For example, for key parameters such as stress and deformation, when the monitoring data exceeds or approaches the preset threshold, an early warning signal is immediately triggered. This is suitable for scenarios with clear risk indicators to ensure real-time monitoring of the key status of the tunnel structure.

[0463] S04.2. The probabilistic risk intelligent assessment algorithm based on Bayesian network integrates the uncertainty factors in the multi-source information set, such as material aging, environmental changes, construction quality, etc., sets the fuzzy set and rules of the tunnel structure health status to build a risk assessment model for the tunnel structure, updates the structural health status data in real time, dynamically calculates the risk probability distribution of the structure, and generates warning signals of different levels, such as low risk, medium risk and high risk. Each level corresponds to different response measures to ensure accurate warning in a complex risk environment;

[0464] S04.3. Combined with the complete set of tunnel structure status information output by the multi-source data fusion module, the deep learning algorithm is used to perform anomaly detection again, and the abnormal patterns in historical data and real-time data are identified by training the model, so as to automatically detect potential risks that may not be identified by traditional algorithms and early warning misjudgments caused by data accumulation errors of structural state information after processing by multiple modules, thereby improving the perception ability of the early warning system for complex and nonlinear risks;

[0465] S04.4. The multi-channel notification system uses IoT technology to connect smart devices such as smart watches, mobile applications, alarms, etc. to transmit warning signals in real time, and designs redundant notification mechanisms to ensure multi-channel coverage. Based on the big data integration platform, the notification system can dynamically evaluate the priority of warning information to ensure that relevant personnel can receive the dynamic response strategy recommendations in a timely manner and process the warning information, thereby improving the timeliness and reliability of notifications;

[0466] S04.5. Keep detailed records of each warning event, including warning time, trigger parameters, over-limit range, response measures and other information, to provide a basis for subsequent analysis, fault tracing and maintenance. Review and analyze the events, evaluate the rationality of threshold settings and the effectiveness of response measures, and feed back the analysis results to the threshold setting module to optimize subsequent warning effects.

[0467] The multi-level warning algorithm in the warning module described in S04.1 includes parameter selection and threshold setting, real-time monitoring and threshold judgment. The specific steps are as follows:

[0468] S04.1.1. The early warning module relies on the multi-source data fusion module to obtain a multi-source information set including the external environment, structural performance status, material characteristics, etc., and input multi-source data. Through these data, the system identifies parameters that are critical to the health status of the tunnel structure, such as stress, strain, crack width, temperature and humidity, displacement, etc.;

[0469] S04.1.2. Calculate the limit values ​​of the tunnel structure under different working conditions by combining the tunnel design specifications, construction standards, and material mechanics and structural mechanics theories. Identify the normal parameter ranges and limit values ​​of the tunnel under different environmental conditions by relying on the historical data comparison and analysis function. Simulate the stress and deformation of the tunnel structure under different loads and environmental conditions by using the numerical analysis and simulation functions to set a reasonable safety threshold.

[0470] S04.1.3. The system compares the real-time collected monitoring data with the pre-set safety threshold. When the real-time value of a key parameter approaches or exceeds the set threshold, the system determines the risk level according to the degree of excess and triggers the corresponding warning signal;

[0471] S04.1.4. Based on the adaptive analysis function of the health status assessment module and the adaptive assessment system data feedback of the early warning system, the system dynamically adjusts the safety thresholds of the key parameters, learns new risk patterns and automatically optimizes the threshold settings to improve the accuracy and timeliness of the early warning.

[0472] The response strategy recommendation module generates specific maintenance and emergency response recommendations based on the real-time analysis results of the fused data. By integrating the historical records and maintenance experience of multi-source data, it provides detailed inspection items, repair methods, and the required tools and materials. At the same time, it provides emergency response plans for high-risk warnings, such as personnel evacuation, traffic control, and emergency reinforcement measures to ensure rapid and effective response in critical situations. The response strategy recommendation module has an adaptive optimization function, which forms a closed-loop management through cyclic learning and data feedback, and continuously adjusts and optimizes the response strategy to adapt to the ever-changing risk situation.

Claims

1. A tunnel structure health monitoring and early warning system based on multi-source twin data fusion, used for real-time monitoring and early warning of the health status of the tunnel structure, characterized by There are four modules in total, including data acquisition module, multi-source data fusion module, health status assessment module, and early warning module. The steps are as follows: S01, data acquisition module: collects health monitoring data of tunnel structure from multiple data sources; S02, multi-source data fusion module: fusing the collected multi-source data; S03, health status assessment module: assessing the health status of the tunnel structure based on fusion data; S04, early warning module: early warning based on health status assessment results; The data acquisition module includes a sensing technology module and a data transmission unit. The sensing technology module is responsible for sensing and collecting key monitoring data of the tunnel structure and the surrounding environment. The monitoring data includes equipment status data of the tunnel, tunnel environment data, traffic status data and various monitoring data of the tunnel structure. The data transmission unit is used to transmit the collected data to the data processing center in real time; The multi-source data fusion module includes a data integration platform module, a data fusion and processing algorithm, and a data correction module, which is used to integrate and process the collected multi-source data; The data integration platform module uniformly manages and coordinates the monitoring data from different sensors and monitoring devices; The data fusion and processing algorithm is used to effectively process and fuse multi-source data and eliminate redundant information and data noise; The data correction is used to detect and correct errors and deviations generated by sensors and monitoring equipment during data collection; The health status assessment module includes intelligent assessment algorithm, multi-level health assessment, historical data comparison and analysis, detailed assessment report generation, and visual display. The intelligent assessment algorithm assesses the health status of the tunnel structure based on fused data and identifies potential tunnel structure problems; Through the multi-level health assessment, the monitoring data and historical data of the tunnel are compared and analyzed in real time to evaluate the change trend of the tunnel structure; The detailed assessment report generation is used to generate a detailed assessment report based on potential tunnel structure problems and tunnel structure change trends, and provide corresponding tunnel maintenance suggestions; The visualization displays the health status and evaluation results of the tunnel structure through charts and 3D models; The early warning module includes a multi-level early warning algorithm, dynamic response strategy recommendations, and a multi-channel notification system. The multi-level early warning algorithm identifies different types and levels of tunnel structural risks based on tunnel health status and assessment results to form early warning results; The dynamic response strategy suggests providing specific maintenance and emergency response suggestions based on different types of warning information; The multi-channel notification system sends warning information to relevant personnel through SMS, email, mobile phone applications and IoT devices; S02.

1. The data integration platform module detects a subset of a multi-source information set according to various sensing technologies in the data acquisition module, wherein the multi-source information set includes a real information set of an external environment, a real information set of a structural performance state, and a real information set of a material characteristic state; S02.2, reducing the dimension of the multi-source information set by dimensionality reduction technology, eliminating redundant information, simplifying the data set structure, and extracting main features, while applying constraints based on known tunnel physical information, and outputting an approximate subset of the multi-source information set; S02.

3. hierarchically process the information of each scale in the approximate subset of the multi-source information set by a multi-scale analysis method, perform precision correction on the information of each scale by combining the spatial resolution, temporal resolution and measurement accuracy of the sensor, and fuse the calibration information at each scale based on the multi-scale reconstruction technology; S02.

4. Estimating the dynamic system in a noisy environment through a recursive algorithm, combining the inherent correlation between each multi-source information, effectively integrating the approximate subset of the multi-source information set, inferring the structural state under certain constraints, and dynamically comparing it with the real information set detected by the sensing technology module, combining prior knowledge and real-time data, making probability estimates and decisions, recursively correcting the constraints, and constructing complete tunnel structure state information; S02.

5. The data correction based on the big data integration platform is used to monitor the quality of multi-source data, evaluate and mark the data, detect and correct errors and deviations generated by sensors and monitoring equipment during the data collection process, automatically identify abnormal data and make corrections, and handle data missing problems caused by equipment failure or data loss by inferring and completing missing data based on historical big data and maintenance records.

2. According to claim 1, a tunnel structure health monitoring and early warning system with multi-source twin data fusion is characterized by: The sensing technology module specifically includes: S01.

1. Monitor the stress and strain of the tunnel structure under different load conditions through stress and strain sensors to capture structural deformation information; S01.

2. Install strain gauges and displacement sensors to record the crack distribution, characteristics and expansion patterns of the tunnel structure, and provide early warning information on potential risks of the structure; S01.

3. Use temperature sensors to monitor temperature changes inside the tunnel and in the surrounding environment, and assess the impact of temperature on tunnel structural materials and overall stability; S01.

4. Monitor the humidity level inside and around the tunnel through humidity sensors to detect corrosion and other potential impacts of the humid environment on the tunnel structure; S01.

5. Monitor the displacement and settlement of the tunnel structure through displacement sensors to ensure the overall stability and long-term safety of the structure; S01.

6. Capture the vibration data of the tunnel structure through vibration sensors and analyze the impact of external environmental factors on the tunnel structure; S01.

7. The geological radar data acquisition unit uses geological radar technology to conduct non-destructive testing of the rock mass and geological environment around the tunnel, obtain geological layers, cracks and water flow information, and ensure the coordination and safety of the tunnel and the surrounding geological environment; The data transmission unit also includes: S01.

8. The data transmission unit transmits the collected data to the data processing center in real time, supporting wired transmission and wireless transmission; S01.

9. The data processing center includes preliminary filtering and preliminary analysis of data, and performs data preprocessing on the collected real-time data.

3. According to claim 2, a tunnel structure health monitoring and early warning system with multi-source twin data fusion is characterized by: The preliminary data filtering includes outlier detection and noise filtering; S01.9.

1. The outlier detection includes: S01.9.

11. Statistical method: For each sensor data, let its observations over a period of time follow a normal distribution, and detect outliers by calculating the z-score of each data point: in x i is the observed value at a certain moment, μ is the mean of historical data, σ is the standard deviation, if z-score If it exceeds the threshold, the data point is considered an outlier; S01.9.12, based on the sliding window method: using the sliding window to calculate the local mean and standard deviation , and detect whether the current data significantly deviates from the statistical characteristics within the window: Where N is the window size of the sliding window, t is the sliding window time point, j is the time index variable, is the observed value at that time point. If the current data x t If the deviation from the mean value within the window exceeds a certain threshold, it is marked as an anomaly; S01.9.

2. The noise filtering includes: S01.9.21, Low-pass filtering: Use a low-pass filter to remove high-frequency noise in the sensor data while retaining the effective components in the signal; Let the sampling frequency be f s , the cutoff frequency of the selected filter f c and filter order n , the transfer function of the low-pass filter for: in w c =2 f c , s is a complex frequency variable, containing information about frequency and attenuation or growth; S01.9.22, Median filtering: Eliminate isolated noise points by replacing the central data point with the median value in the data window: 2 of them b +1 is for window size, is the central data point value after replacement, and median() is the median function; Preliminary data analysis includes data feature extraction and trend analysis; S01.9.3, the data feature extraction includes: S01.9.31, Time Domain Feature Extraction: Calculate Mean and standard deviation , and use skewness and kurtosis to capture the asymmetry and peak degree of the data distribution: Among them, N is the total number of observations, x i is the i-th observation of the data; S01.9.32, Frequency domain feature extraction: Convert time domain data into frequency domain through Fast Fourier Transform FFT to analyze the frequency components of the signal: in is a discrete-time signal, is the time index, yes The Fourier transform of is the frequency index, is an imaginary unit, satisfying j 2 =-1, P is the sequence Length; The power spectral density (PSD) is calculated to identify the main frequency components and their corresponding amplitudes. The PSD is obtained by calculating the autocorrelation function of the signal and performing a Fourier transform (FFT) on it: Where R xx (m) is the autocorrelation function of the signal, and m is the lag of the time delay; S01.9.

4. The trend analysis includes: S01.9.

41. Linear Regression: Analyze time series trends in sensor data using linear regression , estimate the slope and intercept of the trend to determine whether there is a long-term change in the data: in is the intercept, is the slope, is the error term, Represents unexplained random fluctuations or measurement errors in the model; the parameters are estimated using the least squares method and ; S01.9.42 Moving Average: Calculate a moving average of data to smooth short-term fluctuations and show long-term trends: in, is the moving average at time t, is the window size of the moving average, is the observed value at time j.

4. The tunnel structure health monitoring and early warning system with multi-source twin data fusion according to claim 1 is characterized by: In the multi-source information set, different types of data are collected at different time and space scales, and the data need to be time-aligned and spatially interpolated to ensure that different data sources are unified at the same time and space scales; The time alignment includes: resampling data of different frequencies to the same time node by linear interpolation or spline interpolation method; The spatial interpolation includes: when the data sources are at different locations, mapping the data into the same spatial reference frame by the spatial interpolation; Performing dimensionality reduction on the multi-source information set by using dimensionality reduction technology to reduce the dimension, eliminate redundant information, and simplify the data structure; The dimensionality reduction technique is to project high-dimensional data into a low-dimensional subspace while retaining the direction with the largest variance in the data, thereby extracting the most important features; The S02.2 specifically includes: S02.2.

1. Let the multi-source information set be represented as a matrix , where q is the sample size, is the number of features; Each feature is centered and standardized, and the mean is calculated. and standard deviation , convert the original data into a standardized matrix ij : Calculate the covariance matrix : For the covariance matrix Perform eigenvalue decomposition and obtain the eigenvalue and the corresponding eigenvector , the eigenvalue represents the variance of the principal component, and the eigenvector represents the direction of the principal component; Sort by the eigenvalues ​​from large to small, k The largest eigenvalues ​​and their corresponding eigenvectors form a new low-dimensional subspace, which transforms the standardized data matrix Projecting into the low-dimensional subspace; Before selection k feature vector , then the new data is expressed as: in, is a The matrix contains the data after dimensionality reduction; In the process of dimensionality reduction, physical constraints need to be imposed in combination with known physical information of the tunnel to ensure that the results after dimensionality reduction are statistically valid, consistent with the structural characteristics and behavior of the tunnel, and to generate a more accurate approximate subset of the multi-source information set; S02.2.

2. Combine the geometric parameters of the tunnel and take the geometric shape of the tunnel and its position in space as constraints; S02.2.

3. Combined with the material properties of the tunnel: S02.2.

31. Material strength constraints: When evaluating stress and deformation data, if the reduced-dimensional data show that the stress exceeds the yield strength of the material or the deformation exceeds the allowable strain range of the material, mark these data points as abnormal or make corrections to ensure that the reduced-dimensionality results are within a reasonable material range; S02.2.32, Material aging and erosion constraints: When the environmental data shows that certain areas have high erosion or aging, the reduced-dimensional data needs to reflect the performance changes in these areas; S02.2.

4. Combined with structural mechanics constraints: S02.2.41, Stress-strain relationship constraint: According to the mechanical model of the material, ensure that the reduced data maintains the relationship between stress and deformation during the dimensionality reduction process; S02.2.42, Deformation mode constraints: According to the factors and the force mode of the structure, the data is corrected in the dimensionality reduction process to make it conform to the typical tunnel deformation mode. The factors include surface settlement, groundwater changes and vehicle loads during the operation of the tunnel; S02.2.

5. Combining historical data and prior knowledge: Historical trend constraints: By comparing and analyzing historical monitoring data with current data, the dimension reduction process ensures that the results of dimension reduction are consistent with the long-term trend; Prior knowledge constraint: Set a constraint on the data variation range during the dimensionality reduction process so that any principal component data exceeding this range is marked or corrected.

5. The tunnel structure health monitoring and early warning system with multi-source twin data fusion according to claim 1 is characterized by: In S02.4, the approximate subsets of the multi-source information set are effectively fused by combining the intrinsic correlations between the multi-source information, including: S02.4.

1. Model the dynamic system of the tunnel structure and make the state of the tunnel structure Controlled input and external noise The influence can be expressed by the state space model as: in and is the state transfer matrix and control input matrix of the system, is Gaussian white noise, which represents the random disturbance of the system; Output It is expressed as: in is the observation matrix, is the observation noise; S02.4.

2. Estimate the system state via a recursive prediction and update procedure suitable for a linear Gaussian noise environment: S02.4.

21. Initial Setup: Setting the Initial State Estimate and its error covariance matrix : S02.4.

22. Prediction step: Estimation based on current state and input Predict the state at the next time point : And the prediction error covariance matrix : in is the process noise covariance; S02.4.

23. Update step: based on new observation data , update the state estimate: in is the gain matrix: And update the error covariance matrix: in is the observation noise covariance; S02.4.24, Multi-source information fusion: Repeat the above update steps for all sensors 𝑖, and combine the observation results of each sensor, calculate the weighted average or other fusion method, and obtain the final fusion state estimate; S02.4.

3. Based on the intrinsic correlations between different sensor data, all relevant intrinsic correlations are integrated in the fusion algorithm: S02.4.

31. Incorporate the correlation between sensor data into the state space model to make the tunnel stress Affected by temperature The impact of expanding the tunnel state vector To include temperature information, or in the state transfer matrix Introduce temperature-related terms in ; State vector expansion: in is the stress state of the tunnel, is the temperature state of the tunnel; State transition matrix Including the relationship between temperature and stress: Represents the rate of change of the stress state σ(t) itself over time or the autocorrelation term in the transfer relationship, represents the direct effect of temperature on stress, represents a feedback mechanism or interaction, It represents the autocorrelation term in the rate of change or transfer relationship of the temperature state T(t) itself over time; S02.4.

32. Mining the correlation between different sensor data through historical data or real-time data: Representing multi-source data as a matrix X , where each row represents the observation value of a different sensor, and each column represents the observation at a different time point. X Calculate the covariance matrix: Perform eigenvalue decomposition on the covariance matrix and extract the main eigenvectors to reveal the implicit correlation between the sensor data. These principal components will be used as new variables to introduce into the state model. S02.4.

33. Divide the multi-source data into two groups, forming two matrices respectively. X 1 and X 2. Find the linear combination that maximizes the correlation between the two data sets by solving linear algebraic equations: Among them, argmax w1,w2 It means to find the values ​​of w1 and w2 so that the correlation function corr Get the maximum value, 1 and 2 is the weight vector, and represents the transpose of the weight vector, which means that the weight vector and 1 and 2. The result of linear transformation of matrices X1 and X2 is to introduce the obtained related variables into the state model as new state variables; S02.4.

4. During the recursive processing, as new observations arrive, the noise covariance matrix needs to be dynamically adjusted and To adapt to the dynamic changes of the system; S02.4.

41. Process noise covariance matrix : Initial setting through historical data analysis or experience , over time, based on the current state estimation error Make continuous recursive corrections: in α is the smoothing factor, ranging from 0 to 1. Indicated in t Estimate or predict the state of the system at any moment; S02.4.

42. Observation noise covariance matrix : Initial setting based on the noise characteristics of the sensor , in the recursive process, the actual error observed by the sensor is dynamically adjusted : in β is the smoothing factor; S02.4.

43. Adjust the parameters of the state model in S02.4.2 through the feedback mechanism in the recursive calculation so that the model can better adapt to the changes in real-time data; After each update, the state estimation error is calculated And modify the model parameters of the filter: if If the value exceeds the preset threshold range, it means that some parameters of the model need to be adjusted; Use feedback mechanism to dynamically adjust the filter gain : in γ It is the adjustment coefficient. According to the error feedback, the filter parameters are dynamically adjusted to gradually optimize the fusion effect.

6. The tunnel structure health monitoring and early warning system with multi-source twin data fusion according to claim 1 is characterized by: The intelligent assessment algorithm includes a neural network algorithm and a fuzzy logic algorithm, and the multi-level health assessment includes an expert system and numerical analysis; S03.

1. Perform multi-level analysis and processing on the data set processed by the multi-source data fusion module through a neural network algorithm, learn the state distribution characteristics and variable patterns of complex tunnel structures from historical data, and analyze the spatiotemporal evolution laws between data and physical observation phenomena; S03.

2. Establish a mapping relationship between the structural state response and the physical analytical model by combining multiple physical analytical models, and verify its effectiveness through the numerical analysis method. By performing fine meshing and mechanical calculation on the tunnel structure, the stress and deformation of the tunnel structure under various load conditions are simulated to accurately predict the response and potential weak points of the structure; S03.

3. The numerical analysis results constitute a twin data set of the tunnel structure, and the parameters of the physical analytical model are dynamically corrected in combination with the data-driven method to accurately realize the structural state response, correct the mutual feedback of the physical analytical model to improve the data-driven quality, and establish a physical model-data-driven mutual feedback collaborative and dynamic progressive hierarchical analysis mechanism; S03.

4. Using the fuzzy logic algorithm to process the uncertainty and fuzzy information in the output results after the hierarchical analysis mechanism, the information entropy reduction of the structural state space perception is reduced to the greatest extent, and by setting the fuzzy set and rules of the tunnel structure health state, the health level and risk level of the structure are evaluated to provide clear evaluation results; S03.

5. The expert system integrates the knowledge and experience of engineering experts, builds a rule base and a knowledge base, automatically diagnoses and evaluates the health status of the tunnel structure, identifies potential structural problems and failure modes, and systematically compares real-time data with historical data through historical data comparison and analysis, identifies long-term changes and periodic change trends of the tunnel structure and potential degradation problems, and evaluates the life and stability of the structure; S03.

6. Generate a detailed assessment report based on the above assessment results, including structural health score, potential risk location, fault type and specific maintenance recommendations, and provide a detailed description of the fault type and corresponding maintenance recommendations by analyzing the monitoring data and assessment results; S03.

7. The visualization display function uses a variety of charts to intuitively display the structural health score and the trend of changes in monitoring data. At the same time, it uses a 3D model to display the overall situation and evaluation results of the tunnel structure. It supports rotation, zooming and multi-angle viewing, and provides an intuitive display of the structural health status.

7. The tunnel structure health monitoring and early warning system based on multi-source twin data fusion according to claim 6 is characterized in that: The health status assessment module performs multi-level analysis based on the high-quality large data set obtained after processing by the multi-source data fusion module, including: S03.1.

1. Further prepare and organize the good data set processed by the multi-source data fusion module: S03.1.

11. Data Standardization: The data from different sensors Perform standardized min-max scaling and normalization: in is the mean of the data, is the standard deviation of the data; S03.1.12 Data stratification: According to the time span, the data is divided into three levels: short-term, medium-term and long-term. Short-term data is used to capture instantaneous changes, and long-term data is used to reflect trends. Data is layered based on the spatial location of the sensor; S03.1.

2. The multi-level analysis includes analyzing the data at different levels: S03.1.

21. Time hierarchy The changing trend of the short-term data is analyzed by sliding window technology, the moving average is calculated to smooth the data, and the difference method is used to detect the change point: If the change value Δ x t If it exceeds a certain threshold, it is marked as a potential outlier; Use the ARIMA model for long-term trend analysis of time series. The ARIMA model is used to capture the long-term trend and seasonal components of the data: in 1,…, p’ is the autoregressive coefficient, θ 1,…, θ q’ is the moving average coefficient, ,… is the observed value at the current time point, is the error term, c is the constant term, and the optimal value is selected by Akaike Information Criterion AIC. p’ , q’ parameter; S03.1.22, spatial hierarchy By generating a spatial heat map of the sensor data, we visualize the stress, temperature, and humidity distribution of each part of the tunnel structure. We use Moran's I spatial statistics to detect the spatial autocorrelation of the data and determine whether there is any regional abnormal behavior: in is a weight matrix, representing the spatial adjacency relationship; S03.1.

23. Feature Hierarchy Use principal component analysis (PCA) or linear discriminant analysis (LDA) to extract key features and reduce feature dimensions; The data is then analyzed at multiple scales using wavelet transform to decompose different frequency components, where low-frequency components represent the trend of the data and high-frequency components represent instantaneous changes.

8. The tunnel structure health monitoring and early warning system based on multi-source twin data fusion according to claim 1 is characterized in that: The early warning module specifically includes: S04.

1. Use a threshold algorithm to set a safety threshold for the health status of the tunnel structure. When the monitoring data exceeds the preset threshold, an early warning signal is immediately triggered to ensure real-time monitoring of the key status of the tunnel structure; S04.

2. The probabilistic risk intelligent assessment algorithm based on Bayesian network integrates the uncertainty factors in the multi-source information set, sets the fuzzy set and rules of the tunnel structure health status to build a risk assessment model for the tunnel structure, updates the structural health status data in real time, dynamically calculates the risk probability distribution of the structure, and generates warning signals of different levels. Each level corresponds to different response measures, which is used to achieve accurate warning in a complex risk environment; S04.

3. Combine the tunnel structure status information set output by the multi-source data fusion module, use the deep learning algorithm to perform anomaly detection again, identify abnormal patterns in historical data and real-time data through training models, and automatically detect potential risks that are not identified by traditional algorithms and early warning misjudgments caused by data accumulation errors of structure status information after processing by multiple modules; S04.

4. The multi-channel notification system adopts IoT technology to connect smart devices to transmit early warning signals in real time, and designs a redundant notification mechanism to ensure multi-channel coverage. Based on the big data integration platform, the notification system can dynamically evaluate the priority of early warning information to ensure that relevant personnel can receive dynamic response strategy suggestions and process early warning information in a timely manner; S04.

5. Keep a detailed record of each warning event, including the warning time, trigger parameters, limit-exceeding range, and response measures. Review and analyze the events, evaluate the rationality of the threshold setting and the effectiveness of the response measures, and feed back the analysis results to the threshold setting module to optimize the subsequent warning effect.

Images