Digital twin early warning system and method for shield tunnel structure health diagnosis

Abstract

The application discloses a digital twin early warning system and method for shield tunnel structure health diagnosis, belongs to the technical field of tunnel engineering monitoring, and comprises a data acquisition unit, which is composed of point type local high-precision monitoring and distributed wide-area overall monitoring in a complementary manner, and is used for collecting geometric deformation, physical field and apparent state data; a twin model construction unit, which is used for establishing a shield tunnel structure digital twin and dynamically deducing; a health index fusion unit, which is used for fusing multi-source data by using a Bayesian network and a D-S evidence theory, constructing a fused damage index, and generating a comprehensive health index; a time and space prediction unit, which is used for predicting a health index trend; a graded early warning unit, which is used for outputting four-color early warning information according to the comprehensive health index and a predicted value; and a three-dimensional visualization terminal, which is used for displaying a health state in a color cloud chart; the application realizes comprehensive quantification, accurate diagnosis and graded early warning of the health state of the tunnel structure, effectively overcomes the average effect of the spatial resolution of the optical fiber, and improves the damage identification and positioning accuracy.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering monitoring technology, and in particular to a digital twin early warning system and method for structural health diagnosis of shield tunnels. Background Technology

[0002] During the construction and operation phases, the structural health of shield tunnels is highly complex and time-varying due to the coupled influence of multiple factors, including geometric deformation (convergence, settlement), physical fields (stress, strain), and apparent conditions (cracks, water seepage). Accurately perceiving and diagnosing the structural health of tunnels is of great significance for ensuring safe tunnel operation and preventing structural failure.

[0003] The existing shield tunnel health monitoring technology has the following shortcomings: (1) The monitoring system is incomplete, and most of them use a single type of sensor. It lacks the complementary integration of point-type local high-precision monitoring and distributed wide-area overall monitoring, making it difficult to fully capture the global state of the tunnel structure; (2) The spatiotemporal reference of multi-source data is not unified. Data such as geometric deformation, physical field, and apparent state have significant differences in spatial coordinates, sampling frequency, and data structure, making it difficult for traditional methods to achieve effective integration; (3) Digital twin models are mostly statically constructed offline, which cannot dynamically simulate structural deformation based on real-time monitoring data, and lacks the ability to complete and correct missing data.

[0004] Furthermore, sensors such as distributed optical fibers suffer from spatial resolution averaging effects, which can easily distort strain data and smooth out local damage features, resulting in low damage localization accuracy. Existing health diagnosis methods mostly rely on threshold alarms for single physical quantities, lacking comprehensive quantitative indicators that integrate multi-dimensional information, thus failing to achieve graded early warning and proactive intervention. Meanwhile, uncertainties such as the randomness of formation parameters, model errors, and measurement biases are not effectively quantified, leading to a lack of reliability assessment for early warning results.

[0005] Therefore, there is an urgent need for a digital twin system and method for the structural health diagnosis of shield tunnels that can integrate point-based and distributed complementary monitoring, possess dynamic twin simulation capabilities, overcome the average effect of optical fiber spatial resolution, generate a comprehensive health index, and provide graded early warning. Summary of the Invention

[0006] The purpose of this invention is to provide a digital twin early warning system and method for structural health diagnosis of shield tunnels, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides a digital twin early warning system for the structural health diagnosis of shield tunnels: The data acquisition unit is used to collect geometric deformation data, physical field monitoring data, and apparent state data of the shield tunnel. The data acquisition unit includes a point-type local high-precision monitoring subsystem, a distributed wide-area overall monitoring subsystem, and a remote automated acquisition instrument. The point-type local high-precision monitoring subsystem and the distributed wide-area overall monitoring subsystem form a complementary sensing network. The twin model construction unit, connected to the data acquisition unit, is used to establish a digital twin of the shield tunnel structure and map the data collected by the data acquisition unit to the corresponding segment rings of the shield tunnel structure digital twin. The twin model construction unit has a built-in digital twin dynamic simulation model. The digital twin dynamic simulation model dynamically simulates structural deformation by changing the three-dimensional coordinates of the control points and completes and corrects the missing control point data. The health index fusion unit, connected to the twin model construction unit, incorporates a Bayesian network and DS evidence theory model. It is used to perform multi-source fusion of geometric deformation data, physical field monitoring data, and apparent state data. Based on the distributed fiber strain distribution and the second derivative of the distributed fiber strain distribution, it constructs a fusion damage index to generate a comprehensive health index for each segment ring. The spatiotemporal prediction unit, connected to the health index fusion unit, is used to perform time-series prediction of the comprehensive health index of each segment ring, and output the predicted value and evolution trend of the comprehensive health index for future periods. The graded early warning unit, connected to the health indicator fusion unit and the spatiotemporal prediction unit, is used to comprehensively determine the early warning level based on the comprehensive health index and the predicted value of the comprehensive health index for future periods, and generate the risk zone mileage range and handling suggestions corresponding to the early warning level. The 3D visualization terminal, connected to the hierarchical early warning unit, is used to display the health status of the entire tunnel in the form of a color cloud map, and to display detailed information of the early warning segment ring in response to user interaction. The early warning segment ring is the segment ring whose comprehensive health index falls into the early warning threshold range.

[0008] Preferably, the point-type local high-precision monitoring subsystem includes: vibrating wire strain gauges, rebar internal force gauges, piezometers, and pressure sensors; the vibrating wire strain gauges are used to collect the concrete strain of the tunnel segments, the rebar internal force gauges are used to collect the axial force of the rebar in the tunnel segments, the piezometers are used to collect the water pressure in the surrounding rock, and the pressure sensors are used to collect the contact pressure in the surrounding rock; the vibrating wire strain gauges, rebar internal force gauges, piezometers, and pressure sensors are deployed at the monitoring sections and control points determined according to the internal forces calculated based on the shield tunnel structure; a remote automated acquisition instrument is connected to the vibrating wire strain gauges, rebar internal force gauges, piezometers, and pressure sensors; the remote automated acquisition instrument transmits the collected data to a server or database for aggregation and storage via a communication network; The distributed wide-area overall monitoring subsystem includes: distributed fiber optic sensors, laser scanners, and image acquisition equipment. The distributed fiber optic sensors are deployed along the circumferential cross-section and axial survey lines to form a three-dimensional distributed fiber optic sensing network for the tunnel. The distributed fiber optic sensors are used to collect structural strain, temperature, and leakage signals. The laser scanner is used to collect point cloud data of the tunnel inner wall and generate geometric deformation data, including segment ring convergence values, settlement values, and misalignment. The image acquisition equipment is used to collect images of the tunnel inner wall and extract apparent state data through image recognition algorithms. The apparent state data includes crack length, crack width, and the extent of seepage. The physical field monitoring data consists of the concrete strain of the tunnel segment collected by the vibrating wire strain gauge, the axial force of the steel reinforcement of the tunnel segment collected by the steel reinforcement internal force gauge, the water pressure of the surrounding rock collected by the piezometer, the contact pressure of the surrounding rock collected by the pressure sensor, and the structural strain, temperature and leakage signals collected by the distributed optical fiber sensor.

[0009] Preferably, the health indicator fusion unit includes: The indicator tree construction module is used to construct a three-level health indicator tree that includes a geometric deformation indicator layer, a physical field indicator layer, and an apparent state indicator layer. The geometric deformation indicator layer is constructed based on geometric deformation data, the physical field indicator layer is constructed based on physical field monitoring data, and the apparent state indicator layer is constructed based on the apparent state data. The Bayesian network inference module, connected to the indicator tree construction module, is used to calculate the posterior probability of each pipe segment ring under various health states based on the prior probability and conditional probability of each indicator in the three-level health indicator tree. The damage index fusion module, connected to the twin model construction unit, is used to calculate the second derivative of the distributed optical fiber strain distribution to establish strain curvature characteristic parameters, calculate the strain damage probability index based on the distributed optical fiber strain distribution, calculate the strain curvature damage probability index based on the strain curvature characteristic parameters, and fuse the strain damage probability index and the strain curvature damage probability index to generate a fused damage index. The DS evidence theory fusion module, connected to the Bayesian network inference module and the damage index fusion module, is used to generate a comprehensive health index between 0 and 1 by using the posterior probability as the first evidence body and the fused damage index as the second evidence body, after resolving the conflict between the different evidence bodies.

[0010] Preferably, the spatiotemporal prediction unit incorporates a spatiotemporal dual-dimensional coupled prediction model that combines a long short-term memory network and a convolutional neural network, and introduces a spatiotemporal attention mechanism; the spatiotemporal prediction unit takes the comprehensive health index sequence of N consecutive moments before the current moment as input and outputs the predicted comprehensive health index values ​​for the next M moments, where N and M are positive integers.

[0011] Preferably, the graded early warning unit presets a blue threshold range, a yellow threshold range, an orange threshold range, and a red threshold range, which correspond to blue warnings, yellow warnings, orange warnings, and red warnings, respectively; wherein, the upper limit of the red threshold range is lower than the lower limit of the orange threshold range; When the comprehensive health index falls into the blue threshold range, the tiered early warning unit outputs a blue early warning message and records it; when it falls into the yellow threshold range, it outputs a yellow early warning message and a suggestion for attention; when it falls into the orange threshold range, it outputs an orange early warning message and a suggestion for inspection sections; when it falls into the red threshold range, or when the predicted value of the comprehensive health index for a future period will fall into the red threshold range within a preset time, the tiered early warning unit outputs a red early warning message that includes the mileage range of the risk section and the handling measures.

[0012] Preferably, the graded early warning unit includes an uncertainty quantification module. The uncertainty quantification module uses a combination of Bayesian inference and Monte Carlo simulation to quantify the randomness of geological parameters, model errors, and monitoring data deviations in the digital twin of the shield tunnel structure. It outputs a comprehensive health index and the confidence interval and risk probability of the predicted comprehensive health index. The graded early warning unit adjusts the early warning level according to the confidence interval and risk probability.

[0013] A digital twin early warning method for structural health diagnosis of shield tunnels includes the following steps: S1. By combining point-based local high-precision monitoring with distributed wide-area overall monitoring, geometric deformation data, physical field monitoring data and apparent state data of shield tunnels are collected. S2. Establish a digital twin of the shield tunnel structure, which includes geological parameters, structural mechanics model, geometric model and real-time monitoring data mapping; map the collected data to each segment ring corresponding to the digital twin of the shield tunnel structure, and complete and correct the missing control point data through the digital twin dynamic simulation model. S3. Using Bayesian networks and DS evidence theory, multi-source fusion of geometric deformation data, physical field monitoring data and apparent state data is performed. Based on the distributed fiber strain distribution and the second derivative of the distributed fiber strain distribution, a fusion damage index is constructed to generate a comprehensive health index for each segment ring. S4. Taking the comprehensive health index sequence of N consecutive moments before the current moment as input, the spatiotemporal dual-dimensional coupled prediction model combining long short-term memory network and convolutional neural network outputs the predicted value of comprehensive health index for the next M moments. S5. Based on the comprehensive health index and the predicted comprehensive health index for future periods, combined with the uncertainty quantification results, comprehensively determine the warning level, and generate the risk section mileage range and handling suggestions corresponding to the warning level; S6. Display the health status of the entire tunnel in the form of a color cloud map on the 3D visualization terminal, and display detailed information of the warning ring in response to user interaction.

[0014] Preferably, the comprehensive health index generated in S3 for each segment ring includes: S31. Obtain the distributed fiber optic strain distribution acquired by the distributed fiber optic sensor. Calculate the average strain obtained from actual measurement using the following formula. : ; in, The length of the optical fiber spatial resolution. For integration variables; S32. Calculate the absolute value of the second derivative of the average strain to obtain the strain curvature; ; S33, Based on average strain and strain curvature Calculate the strain damage probability index and strain curvature damage probability index ; ; ; in, As the baseline strain under healthy conditions, For strain standard deviation, The baseline curvature under healthy conditions. The standard deviation of curvature; strain damage probability index and strain curvature damage probability index Perform fusion and generate fusion damage indicators; ; in, The weighting coefficients for the strain damage probability index; S34. The posterior probabilities of each health state obtained from Bayesian network inference. The obtained fusion damage index will serve as the primary evidence. As a second piece of evidence; according to the DS evidence theory fusion rules, the degree of trust after fusion is calculated. : S35. Based on the obtained fusion trust level Calculate the comprehensive health index The calculation formula is as follows: ; in, For the first There are three health status levels, including healthy, minor damage, severe damage, and failure. The preset health score, This represents the total number of health status levels.

[0015] Preferably, the specific process by which S34 calculates the fused trust level according to the DS evidence theory fusion rule is as follows: Let the basic probability allocation function of the first piece of evidence be... ,in This indicates that the first body of evidence relates to the state of health. Trust level; Let the basic probability allocation function of the second piece of evidence be: ,in ,when When corresponding to the "damaged" state, This indicates uncertainty about the entire set. The set of all possible health states; According to the DS evidence theory fusion rule, the fused confidence level is given by the following formula: ; in, and These are the hypothetical propositions supported by the first and second pieces of evidence, respectively. To represent the empty set, This represents the conflict coefficient.

[0016] Therefore, the digital twin early warning system and method for structural health diagnosis of shield tunnels described above have the following beneficial effects: (1) This invention constructs a complementary sensing network of point-based local high-precision monitoring and distributed wide-area overall monitoring, and combines three-dimensional spatial distributed optical fiber sensing technology to realize the synchronous acquisition of shield tunnel geometric deformation data, physical field monitoring data and appearance status data. It overcomes the contradiction between accuracy and coverage in a single monitoring method, improves the integrity and redundancy of monitoring data, and has a high sensor survival rate (95% for point-based and 100% for distributed). (2) This invention introduces strain curvature (the second derivative of strain distribution) to construct a fused damage index, which effectively overcomes the masking of damage features by the average effect of optical fiber spatial resolution. By fusing strain damage probability with strain curvature damage probability, the damage features are amplified, independent noise interference is suppressed, and the accuracy of damage identification and localization is improved. (3) Based on Bayesian networks and DS evidence theory, this invention performs multi-source fusion of three types of data: geometric deformation, physical field, and apparent state, and generates a comprehensive health index between 0 and 1, realizing a comprehensive quantitative assessment of structural health status; at the same time, through a preset four-color graded early warning mechanism, differentiated early warning information (blue record, yellow attention, orange inspection, red disposal) can be output according to different threshold ranges of the comprehensive health index, realizing a graded response from minor abnormalities to serious risks; (4) The present invention has a built-in digital twin dynamic simulation model, which can simulate structural deformation in real time by controlling the changes in coordinates of the control points and fill in and correct missing data, thus ensuring the synchronization between the twin model and the physical entity; in addition, the spatiotemporal prediction model combining long short-term memory network and convolutional neural network can predict the evolution trend of health index in advance, provide a time window for preventive maintenance, and output the comprehensive health index prediction value for future periods, thus providing a time window for preventive maintenance. (5) This invention introduces Bayesian inference and Monte Carlo simulation to quantify the randomness of geological parameters, model error and measurement bias, output the confidence interval and risk probability of the comprehensive health index, and adjust the warning level according to the uncertainty, avoiding the false alarm and missed alarm problem of traditional "black and white" warning, and providing a probabilistic and visualized scientific basis for operation and maintenance decision making.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] Figure 1 This is a diagram illustrating the architecture of a digital twin early warning system for structural health diagnosis of shield tunnels, as described in this invention. Figure 2 This is a schematic diagram of the distributed optical fiber arrangement in the circumferential cross-section of a tunnel according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the tunnel axial distributed optical fiber survey line layout according to an embodiment of the present invention; Figure 4 This is a flowchart of a digital twin early warning method for structural health diagnosis of shield tunnels according to the present invention; Figure 5 This is a polar coordinate diagram of circumferential strain monitoring data under different damage states in an embodiment of the present invention; Figure 6 This is a polar coordinate diagram of circumferential strain curvature data under different damage states in an embodiment of the present invention; Figure 7 This is a bar chart showing the distribution of strain damage probability index according to an embodiment of the present invention; Figure 8 This is a bar chart showing the distribution of strain curvature damage probability index according to an embodiment of the present invention; Figure 9This is a bar chart showing the distribution of fused damage indicators based on multi-source information fusion under the P7-damage 1 condition, according to an embodiment of the present invention. Detailed Implementation

[0019] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely illustrates selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0020] Example like Figure 1 As shown, the present invention provides a digital twin early warning system for the structural health diagnosis of shield tunnels, including a data acquisition unit, a twin model construction unit, a health indicator fusion unit, a spatiotemporal prediction unit, a graded early warning unit, and a three-dimensional visualization terminal.

[0021] The data acquisition unit is used to collect geometric deformation data, physical field monitoring data, and apparent state data of the shield tunnel. The data acquisition unit includes a point-type local high-precision monitoring subsystem, a distributed wide-area overall monitoring subsystem, and a remote automated acquisition instrument. The point-type local high-precision monitoring subsystem and the distributed wide-area overall monitoring subsystem form a complementary sensing network. The former provides high-precision, multi-parameter data for key sections, while the latter provides continuous coverage of the entire tunnel. The combination of the two ensures monitoring accuracy and overcomes blind spots.

[0022] The point-based local high-precision monitoring subsystem includes: vibrating wire strain gauges, rebar internal force gauges, piezometers, and pressure sensors. The vibrating wire strain gauges are used to collect the concrete strain of the tunnel segments, with a range of ±1500 με and an accuracy of ±1 με. The rebar internal force gauges are used to collect the axial force of the rebar in the tunnel segments, with a range of ±300 kN. The piezometers are used to collect the water pressure in the surrounding rock, with a range of 0~1 MPa. The pressure sensors are used to collect the contact pressure in the surrounding rock, with a range of 0~2 MPa. These sensors are deployed at monitoring sections and control points determined based on the internal forces calculated according to the shield tunnel structure. In this embodiment, a monitoring section is set every 20 rings (approximately 24 m) along the tunnel longitudinal direction. Each section has four control points at the arch crown, arch bottom, left arch waist, and right arch waist. The collected data is transmitted via a remote automated data acquisition instrument through an RS485 bus or fiber optic communication network to a central server database for aggregation and storage.

[0023] The distributed wide-area overall monitoring subsystem includes: distributed fiber optic sensors, laser scanners, and image acquisition equipment; the distributed fiber optic sensors (using a Brillouin optical time-domain analyzer with a spatial resolution of 0.5m) are deployed along circumferential sections and axial survey lines to form a three-dimensional distributed fiber optic sensing network for the tunnel, such as... Figure 2As shown, in this embodiment, 10 circumferential cross-sections are arranged along the longitudinal direction of the tunnel (located at mileages K0+100, K0+300, K0+500, ..., K1+900, respectively). Distributed optical fibers surround the tunnel once within each circumferential cross-section to collect circumferential structural strain, temperature, and leakage signals. Figure 3 As shown, three longitudinal survey lines are laid out along the tunnel axis (located at the arch crown, left arch waist, and right arch bottom, respectively) to collect the structural strain distribution along the tunnel axis. Figure 2 and Figure 3 Together, they form a three-dimensional spatially distributed fiber optic sensing network for the tunnel. The distributed fiber optic sensors utilize a Brillouin Optical Time Domain Analyzer (BOTDA) with a spatial resolution of 0.5m. A laser scanner (using a Z+F5010X phase-type 3D laser scanner) scans the tunnel interior quarterly, generating point cloud data. This data is processed to obtain geometric deformation data such as segment ring convergence values, settlement values, and misalignment amounts. An image acquisition device (using an industrial area array camera with a resolution of 5 megapixels and equipped with LED supplementary lighting) automatically inspects the tunnel every two weeks, acquiring images of the tunnel interior. A deep learning-based image recognition algorithm (using a Mask R-CNN network) extracts apparent condition data such as crack length, width, and the extent of seepage.

[0024] The physical field monitoring data consists of the concrete strain of the tunnel segment collected by the vibrating wire strain gauge, the axial force of the steel reinforcement of the tunnel segment collected by the steel reinforcement internal force gauge, the water pressure of the surrounding rock collected by the piezometer, the contact pressure of the surrounding rock collected by the pressure sensor, and the structural strain, temperature and leakage signals collected by the distributed optical fiber sensor. The physical field monitoring data centrally reflects the stress state of the tunnel structure and the environmental load.

[0025] The twin model construction unit, connected to the data acquisition unit, is used to create a digital twin of the shield tunnel structure, serving as a precise mapping tool from physical entities to virtual space. The digital twin of the shield tunnel structure includes geological parameters (strata layering, elastic modulus, Poisson's ratio, cohesion, internal friction angle, etc.), a structural mechanics model (beam-spring model or shell model), a geometric model (BIM model), and real-time monitoring data mapping. In this embodiment, the tunnel-soil interaction model is established using ABAQUS finite element software, divided into approximately 500,000 elements. The twin model construction unit maps the data acquired by the data acquisition unit to each segment ring corresponding to the digital twin of the shield tunnel structure—each segment ring (1.2m wide) is bound to its design parameters, material properties, construction records, and real-time monitoring data.

[0026] The twin model construction unit incorporates a digital twin dynamic simulation model. This model dynamically simulates structural deformation through changes in the three-dimensional coordinates of control points: four control points are set for each ring (arch crown, arch base, left arch waist, and right arch waist). Coordinates are measured every two hours using a total station, and the measured coordinate changes are input into the finite element model to drive mesh deformation. For missing control point data (e.g., data loss due to equipment malfunction), Kriging interpolation or Gaussian process regression based on spatial correlation is used for completion and correction. In this embodiment, the data missing rate is controlled within 5%, and the average relative error between the completed data and the measured values ​​is less than 3%.

[0027] The health index fusion unit, connected to the twin model construction unit, incorporates a Bayesian network and DS evidence theory model. It is used to perform multi-source fusion of geometric deformation data, physical field monitoring data, and apparent state data. Based on the distributed fiber strain distribution and its second derivative, it constructs a fused damage index to generate a comprehensive health index for each pipe segment ring. The health index fusion unit is the core analysis engine of the system, solving the key problems of unified quantification and conflict handling of multi-source heterogeneous data.

[0028] The health indicator integration unit includes: The indicator tree construction module is used to construct a three-level health indicator tree, comprising a geometric deformation indicator layer, a physical field indicator layer, and an apparent state indicator layer. The geometric deformation indicator layer is constructed based on geometric deformation data, the physical field indicator layer is constructed based on physical field monitoring data, and the apparent state indicator layer is constructed based on the apparent state data. In this embodiment, the geometric deformation indicator layer includes: segment ring convergence ratio (measured convergence / allowable convergence), differential settlement rate (mm / day), and misalignment (mm); the physical field indicator layer includes: concrete strain ratio (measured strain / ultimate strain), steel reinforcement stress ratio (measured stress / yield stress), and joint opening (mm); the apparent state indicator layer includes: crack width ratio (measured width / allowable width) and seepage area ratio (seepage area / segment area). Each indicator is divided into 5 levels (excellent, good, medium, poor, and critical) according to the health status, and a corresponding membership function is established.

[0029] The Bayesian network inference module, connected to the indicator tree construction module, is used to calculate the posterior probability of each tunnel segment ring under various health states (healthy, slightly damaged, severely damaged, and failed) based on the prior and conditional probabilities of each indicator in the three-level health indicator tree. In this embodiment, the Bayesian network structure is determined by domain experts based on the tunnel defect mechanism, and the conditional probability table is learned using the expectation-maximization algorithm from historical defect data. For example, when the crack width ratio is at the "poor" level, the conditional probability of the tunnel segment being in the "severely damaged" state is 0.75.

[0030] The damage index fusion module, connected to the twin model construction unit, is used to calculate the second derivative of the distributed optical fiber strain distribution to establish strain curvature characteristic parameters, calculate the strain damage probability index based on the distributed optical fiber strain distribution, calculate the strain curvature damage probability index based on the strain curvature characteristic parameters, and fuse the strain damage probability index and the strain curvature damage probability index to generate a fused damage index; it overcomes the smooth masking of local damage features by the optical fiber spatial resolution averaging effect.

[0031] The DS evidence theory fusion module, connected to the Bayesian network inference module and the damage index fusion module, is used to generate a comprehensive health index between 0 and 1 by using the posterior probability as the first evidence body and the fused damage index as the second evidence body, after resolving the conflict between the different evidence bodies.

[0032] The spatiotemporal prediction unit, connected to the health indicator fusion unit, is used to perform time-series prediction of the comprehensive health index of each segment ring, outputting the predicted value and evolution trend of the comprehensive health index for future periods, realizing the leap from "post-event diagnosis" to "pre-event warning". The spatiotemporal prediction unit has a built-in spatiotemporal dual-dimensional coupled prediction model that combines a long short-term memory network and a convolutional neural network, and introduces a spatiotemporal attention mechanism. The spatiotemporal prediction unit takes the comprehensive health index sequence of N consecutive times before the current time as input and outputs the predicted value of the comprehensive health index for the next M times, where N and M are positive integers.

[0033] The tiered early warning unit, connected to the health indicator fusion unit and the spatiotemporal prediction unit, is used to comprehensively determine the early warning level based on the comprehensive health index and the predicted value of the comprehensive health index for future periods, and generate the risk zone mileage range and disposal suggestions corresponding to the early warning level. The tiered early warning unit presets blue threshold intervals, yellow threshold intervals, orange threshold intervals and red threshold intervals, which correspond to blue warning, yellow warning, orange warning and red warning respectively; among them, the upper limit of the red threshold interval is lower than the lower limit of the orange threshold interval.

[0034] When the comprehensive health index falls into the blue threshold range, the graded early warning unit outputs a blue early warning message and records it, indicating that the tunnel structure is in a healthy state and no intervention is required.

[0035] When the value falls within the yellow threshold range, a yellow warning message and suggestions for attention will be output, such as "Strengthen daily inspections and pay attention to the development of cracks".

[0036] When the area falls within the orange threshold range, an orange warning message and a suggested inspection section are output, such as "It is recommended to conduct detailed non-destructive testing on this section and increase the monitoring frequency to once every 6 hours."

[0037] When the comprehensive health index falls into the red threshold range, or when the predicted value of the comprehensive health index for a future period will fall into the red threshold range within a preset time (24 hours in this embodiment), the graded early warning unit outputs red early warning information containing the mileage range of the risk section and the handling measures, such as "The comprehensive health index of the K0+850~K0+920 section is 0.32, and it is predicted to drop to 0.28 after 24 hours. It is recommended to immediately stop construction / operation and take grouting reinforcement or temporary support measures."

[0038] The graded early warning unit includes an uncertainty quantification module. This module uses a combination of Bayesian inference and Monte Carlo simulation to quantify the randomness of geological parameters, model errors, and monitoring data deviations in the digital twin of the shield tunnel structure. It outputs a comprehensive health index and the confidence interval and risk probability of the predicted comprehensive health index. The graded early warning unit adjusts the early warning level according to the confidence interval and risk probability. If the risk probability exceeds 70%, even if the current comprehensive health index is slightly higher than the red threshold, it is upgraded to a red warning.

[0039] A 3D visualization terminal, connected to a tiered early warning unit, displays the overall tunnel health status in a color cloud map format. It also responds to user interaction by displaying detailed information about warning segment rings, which are those whose comprehensive health index falls within the warning threshold range. In this embodiment, the tunnel BIM model is imported using the Unity3D engine. Each segment ring is mapped to a gradient of red (low health) → yellow → green (high health) based on its real-time comprehensive health index. Users can click on any segment ring to display detailed health information, including historical curves of the comprehensive health index, original data for each indicator, warning level, and recommended actions. It also supports timeline scrolling to replay the historical health status evolution. Warning segment rings (i.e., those with a comprehensive health index below 0.6) are highlighted and flashing in the 3D scene, facilitating quick location of risk areas by maintenance personnel.

[0040] like Figure 4 As shown, a digital twin early warning method for structural health diagnosis of shield tunnels includes the following steps: S1. By combining point-based local high-precision monitoring with distributed wide-area overall monitoring, geometric deformation data, physical field monitoring data, and apparent state data of shield tunnels are collected.

[0041] S2. Establish a digital twin of the shield tunnel structure, which includes geological parameters, structural mechanics model, geometric model and real-time monitoring data mapping; map the collected data to each segment ring corresponding to the digital twin of the shield tunnel structure, and complete and correct the missing control point data through the digital twin dynamic simulation model.

[0042] S3. Using Bayesian networks and DS evidence theory, multi-source fusion of geometric deformation data, physical field monitoring data and apparent state data is performed. Based on the distributed fiber strain distribution and the second derivative of the distributed fiber strain distribution, a fusion damage index is constructed to generate a comprehensive health index for each pipe segment ring.

[0043] The comprehensive health index for each segment ring includes: S31. Obtain the distributed fiber optic strain distribution acquired by the distributed fiber optic sensor. Calculate the average strain obtained from actual measurement using the following formula. : ; in, The length of the optical fiber spatial resolution. For integration variables; like Figure 5 As shown, Figure 5 This is a polar coordinate distribution diagram with the circumferential angle of the tunnel segment as the circumferential coordinate and the circumferential strain value as the radial coordinate, which intuitively shows the strain distribution characteristics of the tunnel segment in the circumferential direction under different damage conditions. The circumferential angle (0°~360°) in the polar coordinates corresponds to the circumferential position of the tunnel segment, and the positive and negative values ​​of the radial coordinates correspond to tensile strain and compressive strain, respectively. In the legend, the black curve is the circumferential strain distribution curve under the undamaged state, and the red, orange, and yellow curves correspond to three different damage conditions, namely "P7-Damage 1", "P7-Damage 2", and "P7-Damage 3" (i.e., the first to third damage conditions of the 7th ring segment). The curve distribution shows that the strain curve under the undamaged state has a uniform closed shape; while under the damaged condition, the curve deviates significantly from the undamaged curve at the damage location (approximately 180°~270° circumferential region), and the higher the damage degree, the greater the deviation. This intuitively reflects the strain redistribution phenomenon caused by structural damage, providing the original data basis for subsequent damage index calculations.

[0044] S32. Calculate the absolute value of the second derivative of the average strain to obtain the strain curvature; ; like Figure 6 As shown, a polar coordinate distribution diagram with the circumferential angle of the tube segment as the circumferential coordinate and the circumferential strain curvature value as the radial coordinate is used to characterize the local variation gradient characteristics of the strain distribution; where the circumferential angle (0°~360°) of the polar coordinate corresponds to the circumferential position of the tube segment, and the magnitude of the radial coordinate value characterizes the degree of curvature change in the strain distribution; legend and Figure 5Consistent with each other, the black curve represents the strain curvature distribution under undamaged conditions, while the red, orange, and yellow curves correspond to the "P7-Damage 1 / 2 / 3" conditions, respectively. As the second derivative of strain, strain curvature is more sensitive to changes in local structural stiffness. As can be seen from the figure, the strain curvature curve under undamaged conditions is generally smooth and uniform. However, under damaged conditions, the curve shows a significant curvature abrupt change at the damage location (approximately 180° circumferential region). This abrupt change is more significant than the deviation of the original strain curve, effectively amplifying the signal of local damage and providing key feature data for the subsequent construction of high-sensitivity damage indices.

[0045] S33, Based on average strain and strain curvature Calculate the strain damage probability index and strain curvature damage probability index ; ; ; in, As the baseline strain under healthy conditions, For strain standard deviation, The baseline curvature under healthy conditions. The standard deviation of curvature; like Figure 7 The chart shown is a bar chart with the sample point number as the x-axis and the strain-damage probability index value as the y-axis, used to illustrate the probability distribution characteristics of single-source damage calculated based on strain data. The x-axis, "sample point," represents the serial numbers of discrete monitoring points uniformly distributed along the circumference of the tunnel segment (approximately 55 sample points, corresponding to the full-circumference monitoring locations of the tunnel segment); the y-axis, "strain-damage probability index," is the defined... The value represents the probability of damage at each monitoring location calculated based on strain data; the higher the value, the greater the likelihood of damage at that location. In the legend, cyan bars represent "undamaged points" (most sample points with low and evenly distributed index values), and magenta bars represent "potentially damaged points" (peak sample points with significantly increased index values). As can be seen from the figure, the index values ​​of the sample points in the healthy state are generally at a low level (approximately 0~0.02). However, at the damaged location (sample point number approximately 42), a single significant peak value (approximately 0.08) appears, which can preliminarily locate the damaged location. However, some non-damaged areas still show a small fluctuation in index values, reflecting that the single strain index is affected by noise interference due to changes in the overall force.

[0046] like Figure 8The figure shown is a bar chart with the sample point number as the x-axis and the strain-curvature damage probability index value as the y-axis, used to illustrate the single-source damage probability distribution characteristics calculated based on strain-curvature data; where the x-axis and legend are defined as follows... Figure 7 Consistent, the vertical axis "strain-curvature damage probability index" is defined as The value of the index represents the probability of damage occurrence calculated based on strain curvature data at each monitoring location. As can be seen from the figure, the strain curvature index is significantly more sensitive to local damage than the strain index. In addition to the obvious peak value at the main damage location (sample point number 42), the index value also shows small secondary peak values ​​at other locations. This is because the strain curvature is more sensitive to changes in local stiffness and can capture the minute local deformation features that the strain index cannot reflect. However, the existence of secondary peak values ​​may also introduce the risk of misjudgment, so further verification through multi-source fusion is needed.

[0047] strain damage probability index and strain curvature damage probability index Perform fusion and generate fusion damage indicators; ; in, The weighting coefficients for the strain damage probability index; like Figure 9 The chart shown is a bar chart with the sample point number as the x-axis and the fused damage index value as the y-axis. It displays the distribution characteristics of the comprehensive damage index after multi-source information fusion, corresponding to the "P7-Damage 1" working condition. The x-axis and legend are defined as follows: Figure 7 , Figure 8 Consistent, the vertical axis "fusion damage index" is the defined The result is a weighted fusion of the strain damage probability index and the strain curvature damage probability index. Its value represents the comprehensive damage probability at each monitoring location. As can be seen from the figure, the fused index effectively combines the stability of the strain index with the sensitivity of the strain curvature index, significantly suppressing the noise interference and secondary peaks of the single-source index. Only at the main damage location (approximately sample point number 42) is a single, sharp, significant peak formed. The index value in the undamaged area is close to 0 overall, which greatly improves the accuracy and reliability of damage location and intuitively reflects the beneficial effect of fusing multi-source information.

[0048] S34. The posterior probabilities of each health state obtained from Bayesian network inference. The obtained fusion damage index will serve as the primary evidence. As a second piece of evidence; according to the DS evidence theory fusion rules, the degree of trust after fusion is calculated. : According to the DS evidence theory fusion rule, the specific process for calculating the fused trust level is as follows: Let the basic probability allocation function of the first piece of evidence be... ,in This indicates that the first body of evidence relates to the state of health. Trust level; Let the basic probability allocation function of the second piece of evidence be: ,in ,when When corresponding to the "damaged" state, This indicates uncertainty about the entire set. The set of all possible health states; According to the DS evidence theory fusion rule, the fused confidence level is given by the following formula: ; in, and These are the hypothetical propositions supported by the first and second pieces of evidence, respectively. To represent the empty set, This represents the conflict coefficient.

[0049] S35. Based on the obtained fusion trust level Calculate the comprehensive health index The calculation formula is as follows: ; in, For the first There are three health status levels, including healthy, minor damage, severe damage, and failure. The preset health score, This represents the total number of health status levels.

[0050] S4. Taking the comprehensive health index sequence of N consecutive moments before the current moment as input, the spatiotemporal dual-dimensional coupled prediction model combining long short-term memory network and convolutional neural network outputs the predicted value of comprehensive health index for the next M moments. S5. Based on the comprehensive health index and the predicted comprehensive health index for future periods, combined with the uncertainty quantification results, comprehensively determine the warning level, and generate the risk section mileage range and handling suggestions corresponding to the warning level; S6. Display the health status of the entire tunnel in the form of a color cloud map on the 3D visualization terminal, and display detailed information of the warning ring in response to user interaction.

[0051] Therefore, the present invention adopts the above-mentioned digital twin early warning system and method for the structural health diagnosis of shield tunnels, which realizes the comprehensive quantification, accurate diagnosis and graded early warning of the health status of tunnel structures, effectively overcomes the average effect of optical fiber spatial resolution, and improves the accuracy of damage identification and location.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A digital twin early warning system for structural health diagnosis of shield tunnels, characterized in that, include: The data acquisition unit is used to collect geometric deformation data, physical field monitoring data, and apparent state data of the shield tunnel. The data acquisition unit includes a point-based local high-precision monitoring subsystem, a distributed wide-area overall monitoring subsystem, and a remote automated acquisition instrument. The point-based local high-precision monitoring subsystem and the distributed wide-area overall monitoring subsystem form a complementary sensing network. The twin model construction unit, connected to the data acquisition unit, is used to establish a digital twin of the shield tunnel structure and map the data collected by the data acquisition unit to the corresponding segment rings of the shield tunnel structure digital twin. The twin model construction unit has a built-in digital twin dynamic simulation model. The digital twin dynamic simulation model dynamically simulates structural deformation by changing the three-dimensional coordinates of the control points and completes and corrects the missing control point data. The health index fusion unit, connected to the twin model construction unit, incorporates a Bayesian network and DS evidence theory model. It is used to perform multi-source fusion of geometric deformation data, physical field monitoring data, and apparent state data. Based on the distributed fiber strain distribution and the second derivative of the distributed fiber strain distribution, it constructs a fusion damage index to generate a comprehensive health index for each segment ring. The spatiotemporal prediction unit, connected to the health index fusion unit, is used to perform time-series prediction of the comprehensive health index of each segment ring, and output the predicted value and evolution trend of the comprehensive health index for future periods. The graded early warning unit, connected to the health indicator fusion unit and the spatiotemporal prediction unit, is used to comprehensively determine the early warning level based on the comprehensive health index and the predicted value of the comprehensive health index for future periods, and generate the risk zone mileage range and handling suggestions corresponding to the early warning level. The 3D visualization terminal, connected to the hierarchical early warning unit, is used to display the health status of the entire tunnel in the form of a color cloud map, and to display detailed information of the early warning segment ring in response to user interaction. The early warning segment ring is the segment ring whose comprehensive health index falls into the early warning threshold range.

2. The digital twin early warning system for structural health diagnosis of shield tunnels according to claim 1, characterized in that: The point-based local high-precision monitoring subsystem includes: vibrating wire strain gauges, rebar internal force gauges, piezometers, and pressure sensors. The vibrating wire strain gauges are used to collect the concrete strain of the tunnel segments, the rebar internal force gauges are used to collect the axial force of the rebar in the tunnel segments, the piezometers are used to collect the water pressure in the surrounding rock, and the pressure sensors are used to collect the contact pressure in the surrounding rock. The vibrating wire strain gauges, rebar internal force gauges, piezometers, and pressure sensors are deployed at the monitoring sections and control points determined based on the internal forces calculated according to the shield tunnel structure. The remote automated acquisition instrument is connected to the vibrating wire strain gauges, rebar internal force gauges, piezometers, and pressure sensors. The remote automated acquisition instrument transmits the collected data to a server or database for aggregation and storage via a communication network. The distributed wide-area overall monitoring subsystem includes: distributed fiber optic sensors, laser scanners, and image acquisition equipment. The distributed fiber optic sensors are deployed along the circumferential cross-section and axial survey lines to form a three-dimensional distributed fiber optic sensing network for the tunnel. The distributed fiber optic sensors are used to collect structural strain, temperature, and leakage signals. The laser scanner is used to collect point cloud data of the tunnel inner wall and generate geometric deformation data, including segment ring convergence values, settlement values, and misalignment. The image acquisition equipment is used to collect images of the tunnel inner wall and extract apparent state data through image recognition algorithms. The apparent state data includes crack length, crack width, and the extent of seepage. The physical field monitoring data consists of the concrete strain of the tunnel segment collected by the vibrating wire strain gauge, the axial force of the steel reinforcement of the tunnel segment collected by the steel reinforcement internal force gauge, the water pressure of the surrounding rock collected by the piezometer, the contact pressure of the surrounding rock collected by the pressure sensor, and the structural strain, temperature and leakage signals collected by the distributed optical fiber sensor.

3. The digital twin early warning system for structural health diagnosis of shield tunnels according to claim 1, characterized in that, The health indicator integration unit includes: The indicator tree construction module is used to construct a three-level health indicator tree that includes a geometric deformation indicator layer, a physical field indicator layer, and an apparent state indicator layer. The geometric deformation indicator layer is constructed based on geometric deformation data, the physical field indicator layer is constructed based on physical field monitoring data, and the apparent state indicator layer is constructed based on the apparent state data. The Bayesian network inference module, connected to the indicator tree construction module, is used to calculate the posterior probability of each pipe segment ring under various health states based on the prior probability and conditional probability of each indicator in the three-level health indicator tree. The damage index fusion module, connected to the twin model construction unit, is used to calculate the second derivative of the distributed optical fiber strain distribution to establish strain curvature characteristic parameters, calculate the strain damage probability index based on the distributed optical fiber strain distribution, calculate the strain curvature damage probability index based on the strain curvature characteristic parameters, and fuse the strain damage probability index and the strain curvature damage probability index to generate a fused damage index. The DS evidence theory fusion module, connected to the Bayesian network inference module and the damage index fusion module, is used to generate a comprehensive health index between 0 and 1 by using the posterior probability as the first evidence body and the fused damage index as the second evidence body, after resolving the conflict between the different evidence bodies.

4. A digital twin early warning system for structural health diagnosis of shield tunnels according to claim 1, characterized in that: The spatiotemporal prediction unit incorporates a spatiotemporal dual-dimensional coupled prediction model that combines a long short-term memory network and a convolutional neural network, and introduces a spatiotemporal attention mechanism. The spatiotemporal prediction unit takes the comprehensive health index sequence of N consecutive moments before the current moment as input and outputs the predicted comprehensive health index values ​​for the next M moments, where N and M are positive integers.

5. A digital twin early warning system for structural health diagnosis of shield tunnels according to claim 1, characterized in that: The tiered early warning unit is pre-defined with blue, yellow, orange, and red threshold ranges, corresponding to blue, yellow, orange, and red alerts, respectively; the upper limit of the red threshold range is lower than the lower limit of the orange threshold range. When the comprehensive health index falls into the blue threshold range, the tiered early warning unit outputs a blue early warning message and records it; when it falls into the yellow threshold range, it outputs a yellow early warning message and a suggestion for attention; when it falls into the orange threshold range, it outputs an orange early warning message and a suggestion for inspection sections; when it falls into the red threshold range, or when the predicted value of the comprehensive health index for a future period will fall into the red threshold range within a preset time, the tiered early warning unit outputs a red early warning message that includes the mileage range of the risk section and the handling measures.

6. A digital twin early warning system for structural health diagnosis of shield tunnels according to claim 1, characterized in that: The graded early warning unit includes an uncertainty quantification module. This module uses a combination of Bayesian inference and Monte Carlo simulation to quantify the randomness of geological parameters, model errors, and monitoring data deviations in the digital twin of the shield tunnel structure. It outputs a comprehensive health index and the confidence interval and risk probability of the predicted comprehensive health index. The graded early warning unit adjusts the early warning level based on the confidence interval and risk probability.

7. A digital twin early warning method for structural health diagnosis of shield tunnels, applied to a digital twin early warning system for structural health diagnosis of shield tunnels as described in any one of claims 1-6, characterized in that, Includes the following steps: S1. By combining point-based local high-precision monitoring with distributed wide-area overall monitoring, geometric deformation data, physical field monitoring data and apparent state data of shield tunnels are collected. S2. Establish a digital twin of the shield tunnel structure, which includes geological parameters, structural mechanics model, geometric model and real-time monitoring data mapping; map the collected data to each segment ring corresponding to the digital twin of the shield tunnel structure, and complete and correct the missing control point data through the digital twin dynamic simulation model. S3. Using Bayesian networks and DS evidence theory, multi-source fusion of geometric deformation data, physical field monitoring data and apparent state data is performed. Based on the distributed fiber strain distribution and the second derivative of the distributed fiber strain distribution, a fusion damage index is constructed to generate a comprehensive health index for each segment ring. S4. Taking the comprehensive health index sequence of N consecutive moments before the current moment as input, the spatiotemporal dual-dimensional coupled prediction model combining long short-term memory network and convolutional neural network outputs the predicted value of comprehensive health index for the next M moments. S5. Based on the comprehensive health index and the predicted comprehensive health index for future periods, combined with the uncertainty quantification results, comprehensively determine the warning level, and generate the risk section mileage range and handling suggestions corresponding to the warning level; S6. Display the health status of the entire tunnel in the form of a color cloud map on the 3D visualization terminal, and display detailed information of the warning ring in response to user interaction.

8. The digital twin early warning method for structural health diagnosis of shield tunnels according to claim 7, characterized in that, The comprehensive health index of each segment ring generated in S3 includes: S31. Obtain the distributed fiber optic strain distribution acquired by the distributed fiber optic sensor. Calculate the average strain obtained from actual measurement using the following formula. : ； in, The length of the optical fiber spatial resolution. For integration variables; S32. Calculate the absolute value of the second derivative of the average strain to obtain the strain curvature; ； S33, Based on average strain and strain curvature Calculate the strain damage probability index and strain curvature damage probability index ; ； ； in, As the baseline strain under healthy conditions, For strain standard deviation, The baseline curvature under healthy conditions. The standard deviation of curvature; strain damage probability index and strain curvature damage probability index Perform fusion and generate fusion damage indicators; ； in, The weighting coefficients for the strain damage probability index; S34. The posterior probabilities of each health state obtained from Bayesian network inference. The obtained fusion damage index will serve as the primary evidence. As a second piece of evidence; according to the DS evidence theory fusion rules, the degree of trust after fusion is calculated. : S35. Based on the obtained fusion trust level Calculate the comprehensive health index The calculation formula is as follows: ； in, For the first There are three health status levels, including healthy, minor damage, severe damage, and failure. The preset health score, This represents the total number of health status levels.

9. A digital twin early warning method for structural health diagnosis of shield tunnels according to claim 8, characterized in that, S34 The specific process for calculating the fused trust level according to the DS evidence theory fusion rule is as follows: Let the basic probability allocation function of the first piece of evidence be... ,in This indicates that the first body of evidence relates to the state of health. Trust level; Let the basic probability allocation function of the second piece of evidence be: ,in ,when When corresponding to the "damaged" state, This indicates uncertainty about the entire set. It is the set of all possible health states; According to the DS evidence theory fusion rule, the fused confidence level is given by the following formula: ； in, and These are the hypothetical propositions supported by the first and second pieces of evidence, respectively. To represent the empty set, This represents the conflict coefficient.

Images