Shield tunnel deformation and leakage combined monitoring method based on distributed optical fiber sensing

Abstract

The application belongs to the technical field of remote monitoring, and particularly relates to a shield tunnel deformation and leakage combined monitoring method based on distributed optical fiber sensing, which comprises the following steps: step one: arranging an ultra-weak optical grating array sensing optical fiber along the inner wall of a segment of the shield tunnel, and recording the modulated light intensity value by a photoelectric detector to form a compressed sampling vector; step two: dividing the ultra-weak optical grating array sensing optical fiber into a plurality of sensing segments, and obtaining a strain distribution curve and a temperature distribution curve along the shield tunnel according to a reconstructed wavelength distribution curve; step three: respectively extracting strain abnormal features and temperature abnormal features according to the strain distribution curve and the temperature distribution curve; and step four: generating corresponding grade alarm information according to each abnormal mark obtained in step three. The application realizes high-density continuous monitoring of the full length of the shield tunnel, and the compressed sampling mode reduces the data acquisition complexity and improves the resolution speed.

Description

Technical Field

[0001] This invention belongs to the field of remote monitoring technology, specifically relating to a method for joint monitoring of deformation and leakage in shield tunnels based on distributed optical fiber sensing. Background Technology

[0002] Traditional tunnel deformation monitoring relies primarily on manual inspections and periodic measurements, using instruments such as total stations and levels to observe tunnel cross-sectional convergence and longitudinal settlement. While this method offers high accuracy, it has significant limitations: low measurement frequency (typically monthly or quarterly) makes it difficult to capture rapidly evolving deformations; a limited number of measuring points only provides discrete deformation information, failing to comprehensively reflect the structural state along the entire tunnel length; and the need for fixed measuring points within the tunnel and manual operation significantly restricts its effectiveness due to operational constraints. In recent years, automated total stations and hydrostatic leveling systems have been increasingly applied to tunnel monitoring, enabling continuous automatic observation of some points, but these still cannot solve the problems of sparse measuring points and limited coverage. Tunnel leakage monitoring also faces technological bottlenecks.

[0003] Traditional leakage detection relies primarily on manual visual inspection, observing signs such as water stains and calcification deposits on the tunnel lining surface to determine the location of leaks. This method is heavily influenced by human factors and can only detect leaks that have progressed to a certain stage, making early warning difficult. Infrared thermal imaging technology can identify leakage areas by detecting surface temperature differences, but it is greatly affected by ambient temperature fluctuations and requires professionals to carry equipment into the tunnel for scanning, making continuous online monitoring difficult. Geophysical methods such as resistivity and spontaneous potential have also been attempted for leakage detection, but these methods are poorly adapted to the complex metal lining environment of shield tunnels, limiting their practical effectiveness.

[0004] The development of distributed fiber optic sensing technology has provided a new technical approach for monitoring the health of tunnel structures. Distributed fiber optic sensing technology based on Brillouin scattering can continuously measure strain and temperature distribution along the entire length of the fiber, with a single fiber covering a monitoring range of tens of kilometers, making it very suitable for monitoring linear structures like tunnels. However, the spatial resolution of Brillouin optical time-domain analysis is typically around 1 meter, and each measurement takes a long time, making it difficult to meet the monitoring requirements of large-scale, dense measurement points and rapid response. Fiber Bragg grating sensing technology has high measurement accuracy and response speed, but traditional fiber Bragg gratings use wavelength division multiplexing for demodulation. The number of gratings that can be connected in series on a single fiber is limited by the light source bandwidth and grating wavelength spacing, typically not exceeding a few dozen, making it difficult to achieve long-distance, high-density distributed monitoring. Summary of the Invention

[0005] Therefore, the main objective of this invention is to provide a method for joint monitoring of deformation and leakage in shield tunnels based on distributed optical fiber sensing. This method achieves high-density continuous monitoring of the entire length of the shield tunnel. The compressed sampling method reduces the complexity of data acquisition and improves demodulation speed. The segmented cascade reconstruction method makes full use of the tunnel's structural features to improve computational efficiency and reconstruction accuracy. The joint judgment method can identify the coupling relationship between deformation and leakage. The hierarchical alarm and inspection route generation provide scientific decision support for maintenance personnel.

[0006] The technical solution adopted in this invention is as follows: A method for joint monitoring of shield tunnel deformation and leakage based on distributed optical fiber sensing includes the following steps: Step 1: Lay ultra-weak grating array sensing optical fibers along the inner wall of the shield tunnel segments. Use a broadband light source to input detection light into the ultra-weak grating array sensing optical fibers. After the reflected spectrum is wavelength-expanded by a dispersive element, it is projected onto a spatial light modulator. A random modulation pattern is loaded on the spatial light modulator to modulate the wavelength-expanded light signal. The photodetector records the modulated light intensity value to form a compressed sampling vector. Step 2: Divide the ultra-weak grating array sensing fiber into several sensing segments, perform segmented compressed sensing reconstruction on the compressed sampling vector to obtain the reconstructed wavelength sequence of each sensing segment, perform boundary continuity correction on the reconstructed wavelength sequences of adjacent sensing segments, and splice the reconstructed wavelength sequences of each sensing segment after correction to form a reconstructed wavelength distribution curve. Based on the reconstructed wavelength distribution curve, calculate the strain distribution curve and temperature distribution curve distributed along the shield tunnel. Step 3: Extract strain anomaly features and temperature anomaly features from the strain distribution curve and temperature distribution curve respectively, and mark deformation anomaly, leakage anomaly, or leakage-deformation coupling anomaly for each section of the shield tunnel based on the strain anomaly features and temperature anomaly features. Step 4: Generate alarm information of corresponding levels based on the anomaly markers obtained in Step 3, and generate inspection and verification routes based on the level and spatial location of each alarm information.

[0007] Furthermore, the ultra-weak grating array sensing fiber contains several ultra-weak fiber Bragg gratings arranged in series, and the reflectivity of each ultra-weak fiber Bragg grating is set to be between 0.01% and 0.1%.

[0008] Furthermore, the spatial light modulator is a digital micromirror array, and the random modulation pattern is a Bernoulli random binary pattern. Each micromirror element in the Bernoulli random binary pattern is independently set to an on or off state with a 50% probability. The micromirror element in the on state reflects the incident light to the photodetector, and the micromirror element in the off state deflects the incident light to the absorption end. Several different Bernoulli random binary patterns are loaded sequentially. The photodetector records the light intensity integral value once after each Bernoulli random binary pattern is loaded. All the light intensity integral values ​​are arranged in the loading order to form a compressed sampling vector.

[0009] Furthermore, in step two, the ultra-weak fiber Bragg array sensing fiber is divided into several sensing segments according to the segment ring structure of the shield tunnel. Each sensing segment corresponds to a segment ring and includes all ultra-weak fiber Bragg gratings within the corresponding segment ring range.

[0010] Furthermore, in step two, the segmented compressed sensing reconstruction includes: for each sensing segment, extracting column vectors from the random modulation pattern corresponding to the wavelength expansion positions of each ultra-weak fiber Bragg grating within the current sensing segment; arranging the extracted column vectors according to the pattern loading order to form the local observation matrix of the current sensing segment; separating the local compressed sampling sub-vectors corresponding to the current sensing segment from the compressed sampling vectors; and performing iterative hard threshold reconstruction for each sensing segment. The execution process of iterative hard threshold reconstruction is as follows: initializing the center wavelength of each ultra-weak fiber Bragg grating within the current sensing segment to the design nominal wavelength of each ultra-weak fiber Bragg grating; using discrete cosine transform as the sparse representation basis, calculating the current wavelength... The long estimate is transformed by the coefficients under the sparse representation basis. The top few coefficients with the largest absolute value are retained and the remaining coefficients are set to zero. The inverse discrete cosine transform is performed on the retained coefficients to obtain the updated wavelength estimate. The difference between the product of the local observation matrix and the updated wavelength estimate and the local compressed sampling subvector is calculated as the residual vector. The residual vector is backpropagated along the transpose of the local observation matrix and superimposed on the current wavelength estimate to form the corrected wavelength estimate. The process of transformation coefficient calculation, threshold retention, inverse transformation, residual calculation and backpropagation correction is repeated until the L2 norm of the residual vector is less than the preset residual convergence threshold. The reconstructed wavelength sequence of the current sensing segment is then output.

[0011] Furthermore, in step two, the boundary continuity correction includes: at the boundary between two adjacent sensing segments, extracting several wavelength values ​​located at the end of the reconstructed wavelength sequence of the previous sensing segment on the boundary side and several wavelength values ​​located at the beginning of the reconstructed wavelength sequence of the subsequent sensing segment on the boundary side; calculating the boundary jump variable between the linear extrapolated predicted value of the several wavelength values ​​at the end and the several wavelength values ​​at the beginning, and calculating the reverse boundary jump variable between the linear backward extrapolated predicted value of the several wavelength values ​​at the beginning and the several wavelength values ​​at the end; and using half of the boundary jump variable as the boundary jump variable for the previous sensing segment. The boundary correction for each segment is calculated by using half of the reverse boundary jump variable as the boundary correction for the next sensing segment. For each wavelength value in the reconstructed wavelength sequence of the previous sensing segment, the boundary correction is applied in a linearly decreasing manner according to the grating index position of each wavelength value from the boundary. The wavelength value closest to the boundary is given a full boundary correction, the wavelength value farthest from the boundary is given a zero correction, and the correction applied to the wavelength value in the middle position is determined by linear interpolation according to the position ratio from the boundary. The boundary correction for the next sensing segment is applied in the same manner.

[0012] Furthermore, in step two, after performing boundary continuity correction, the reconstructed wavelength sequence of each sensing segment after correction is used as the new initial value, and iterative hard threshold reconstruction is performed again for each sensing segment. When performing iterative hard threshold reconstruction again, the sparse representation basis is replaced with a local Fourier basis constructed with the reconstructed wavelength sequence after correction. The frequency range of each basis vector of the local Fourier basis is limited to the typical frequency range of wavelength changes caused by the deformation and leakage of the shield tunnel structure.

[0013] Furthermore, in step two, the process of obtaining the strain distribution curve and temperature distribution curve based on the reconstructed wavelength distribution curve includes: setting several temperature reference gratings at intervals along the deployment path of the ultra-weak grating array sensing fiber; the temperature reference gratings are encapsulated with loose tubes to eliminate strain transmission so that the temperature reference gratings only respond to temperature changes; extracting the wavelength values ​​of each temperature reference grating position from the reconstructed wavelength distribution curve; calculating the temperature value of each temperature reference grating position based on the offset of the wavelength value of each temperature reference grating position relative to the nominal wavelength and the pre-calibrated temperature sensitivity coefficient; obtaining the temperature distribution curve by linear interpolation for the segments between adjacent temperature reference gratings; subtracting the wavelength offset component caused by temperature determined by the temperature distribution curve from the wavelength offset at each position in the reconstructed wavelength distribution curve to obtain the wavelength offset caused by pure strain; and converting the wavelength offset caused by pure strain into a strain distribution curve based on the pre-calibrated strain sensitivity coefficient.

[0014] Furthermore, in step three, the joint determination of deformation and leakage includes: traversing the strain distribution curve and temperature distribution curve along the longitudinal direction of the shield tunnel using a sliding window of fixed length; calculating the difference between the maximum and minimum strain values ​​within each sliding window as the strain amplitude index for each sliding window; calculating the decrease in temperature value within each sliding window relative to the average temperature of adjacent sliding windows as the temperature negative bias index for each sliding window; when the strain amplitude index of a certain sliding window exceeds a preset strain anomaly threshold, the corresponding sliding window is marked as a deformation anomaly window; when the temperature negative bias index of a certain sliding window exceeds a preset temperature anomaly threshold, the corresponding sliding window is marked as a suspected leakage window; when a sliding window is simultaneously marked as both a deformation anomaly window and a suspected leakage window, the corresponding sliding window is marked as a leakage-deformation coupling window.

[0015] Furthermore, in step four, the leakage-deformation coupling window generates a level one alarm, the suspected leakage window generates a level two alarm, and the abnormal deformation window generates a level three alarm. All sliding windows that generate alarms are sorted in descending order of alarm level, and the tunnel mileage positions corresponding to each sliding window that generates alarms are connected in sequence to generate an inspection and verification route and output to the monitoring terminal.

[0016] By adopting the above technical solutions, this invention achieves the following beneficial effects: Through the combination of ultra-weak grating array sensing fiber and compressed sensing demodulation technology, high-density continuous monitoring of the entire length of the shield tunnel is realized, overcoming the limitations of traditional monitoring methods such as sparse measuring points and limited coverage. The numerous ultra-weak fiber Bragg gratings connected in series in the ultra-weak grating array sensing fiber can be densely distributed along the tunnel's longitudinal direction, with a single fiber covering a monitoring range of several kilometers, significantly reducing the complexity and cost of sensor link deployment. The compressed sampling method, employing a digital micromirror array loaded with a Bernoulli random binary pattern for spatial modulation, significantly reduces the number of sampling measurements required, lowering the complexity and cost of data acquisition hardware, while simultaneously improving the system's demodulation speed and meeting the timeliness requirements of real-time monitoring.

[0017] The proposed method for segmented cascaded compressed sensing reconstruction based on segment ring constraints in shield tunnels fully utilizes the physical characteristics of the shield tunnel segment ring structure. It decomposes the global reconstruction problem into multiple local reconstruction problems for parallel processing, effectively reducing computational complexity and improving reconstruction efficiency. By using cascaded boundary continuity correction, it eliminates the artificial discontinuities caused by independent segment reconstruction, ensuring the smooth continuity of the reconstructed wavelength distribution curve across the entire tunnel. The secondary iterative refinement reconstruction employs a specially designed local Fourier basis, further improving reconstruction accuracy and ensuring that the final strain and temperature distribution curves accurately reflect the true state of the tunnel structure. The use of a temperature reference grating enables effective separation of strain and temperature calculations, avoiding interference from their cross-sensitivity on the monitoring results. Attached Figure Description

[0018] Figure 1 A schematic diagram of the principle and physical architecture of the ultra-weak grating demodulation optical path based on digital micromirror array compressed sampling provided in this embodiment of the invention; Figure 2 A schematic diagram of the reconstructed wavelength distribution curve along the entire length of the shield tunnel provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the temperature distribution curve along the shield tunnel provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the sparse coefficient distribution of the ultra-weak fiber Bragg grating wavelength signal in a single sensing segment in the discrete cosine transform domain and the hard threshold truncation process provided in an embodiment of the present invention. Detailed Implementation

[0019] Any feature disclosed in this specification (including any appended claims and abstract) may be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0020] A method for joint monitoring of shield tunnel deformation and leakage based on distributed optical fiber sensing includes the following steps: Step 1: Lay ultra-weak grating array sensing optical fibers along the inner wall of the shield tunnel segments. Use a broadband light source to input detection light into the ultra-weak grating array sensing optical fibers. After the reflected spectrum is wavelength-expanded by a dispersive element, it is projected onto a spatial light modulator. A random modulation pattern is loaded on the spatial light modulator to modulate the wavelength-expanded light signal. The photodetector records the modulated light intensity value to form a compressed sampling vector. Step 2: Divide the ultra-weak grating array sensing fiber into several sensing segments, perform segmented compressed sensing reconstruction on the compressed sampling vector to obtain the reconstructed wavelength sequence of each sensing segment, perform boundary continuity correction on the reconstructed wavelength sequences of adjacent sensing segments, and splice the reconstructed wavelength sequences of each sensing segment after correction to form a reconstructed wavelength distribution curve. Based on the reconstructed wavelength distribution curve, calculate the strain distribution curve and temperature distribution curve distributed along the shield tunnel. Step 3: Extract strain anomaly features and temperature anomaly features from the strain distribution curve and temperature distribution curve respectively, and mark deformation anomaly, leakage anomaly, or leakage-deformation coupling anomaly for each section of the shield tunnel based on the strain anomaly features and temperature anomaly features. Step 4: Generate alarm information of corresponding levels based on the anomaly markers obtained in Step 3, and generate inspection and verification routes based on the level and spatial location of each alarm information.

[0021] During long-term operation, shield tunnels are affected by various factors such as ground subsidence, groundwater seepage, and train vibration, which can lead to deformation and even leakage of the tunnel segments. In order to achieve comprehensive monitoring of the structural health status of shield tunnels, this embodiment uses ultra-weak grating array sensing optical fibers as distributed sensing elements, which are continuously deployed along the longitudinal direction of the tunnel to form a sensing link covering the entire tunnel length.

[0022] The ultra-weak fiber grating array sensing fiber is composed of several ultra-weak fiber Bragg gratings connected in series. These ultra-weak fiber Bragg gratings are sequentially inscribed along the fiber axis at fixed intervals. Compared with traditional fiber Bragg gratings, ultra-weak fiber Bragg gratings have extremely low reflectivity, typically set in the range of 0.01% to 0.1%. This low reflectivity design has significant technical implications: when a large number of gratings are connected in series in the fiber, each grating reflects only a very small amount of incident light energy, with the vast majority of the light energy continuing to propagate forward and being reflected sequentially by subsequent gratings. This allows thousands or even tens of thousands of gratings to be connected in series on a single fiber without a sharp attenuation of light energy due to cumulative reflection. In one specific embodiment, the ultra-weak fiber grating array sensing fiber contains 5000 ultra-weak fiber Bragg gratings, with a spacing of 0.5 meters between adjacent gratings, thereby achieving continuous coverage monitoring of a 2.5-kilometer-long tunnel.

[0023] When deploying ultra-weak grating array sensing fibers, the fibers need to be laid along the inner wall of the tunnel segments. Several deployment methods can be used. A typical method is to lay them longitudinally along the tunnel arch, a critical area of ​​stress concentration and deformation sensitivity in the tunnel structure. Deploying sensing fibers here can effectively capture structural deformation signals. Another option is to lay multiple sensing fibers simultaneously along the tunnel sidewalls and floor, forming a three-dimensional monitoring network around the tunnel cross-section. This method can obtain more comprehensive structural deformation information, but it increases system complexity and cost. The fibers can be fixed to the inner wall of the tunnel segments using epoxy resin bonding or pre-embedded cable channels. Epoxy resin bonding is simple to install but has slightly lower long-term stability. Pre-embedded channels require reserving installation channels during the segment prefabrication stage, which has higher construction requirements but provides better long-term reliability for the sensing fibers.

[0024] After the sensing fiber optic cable is deployed, the system uses a broadband light source to input probe light into the ultra-weak fiber Bragg array sensing fiber. The light emitted by the broadband light source has a wide wavelength range, capable of covering the reflection wavelength range of all ultra-weak fiber Bragg gratings in the ultra-weak fiber Bragg array sensing fiber. In one specific embodiment, the broadband light source is an amplified spontaneous emission source, with a center wavelength of 1550 nm, a 3 dB bandwidth of 40 nm, and an output optical power of 10 mW. Compared to superluminescent diodes, the amplified spontaneous emission source has a flatter output spectrum, ensuring that the probe light intensity received by each ultra-weak fiber Bragg grating is basically consistent, thereby improving the accuracy of subsequent signal demodulation.

[0025] When broadband probe light enters the sensing fiber of an ultra-weak fiber Bragg grating array, each ultra-weak fiber Bragg grating reflects light of a specific wavelength matching its grating period; this specific wavelength is called the Bragg wavelength. The Bragg wavelength is directly related to the grating period and the effective refractive index of the fiber core. When strain or temperature changes occur at the location of the grating, both the grating period and the effective refractive index change accordingly, causing a corresponding shift in the Bragg wavelength. By detecting the Bragg wavelength shift of each ultra-weak fiber Bragg grating, the strain and temperature changes at the grating location can be deduced. Because the sensing fiber of the ultra-weak grating array contains a large number of ultra-weak fiber Bragg gratings, the light reflected by each grating is superimposed in the fiber to form a composite reflection spectrum, which carries the wavelength information of all the gratings.

[0026] After returning from the sensing fiber of the ultra-weak grating array, the reflected spectrum is guided through a fiber optic circulator to a dispersive element for wavelength unfolding. The function of the dispersive element is to separate light of different wavelengths in space, converting the spectral information that was originally mixed in the temporal domain into spatial distribution information. In one specific embodiment, the dispersive element is a transmission diffraction grating with a grating line density of 1200 lines per millimeter. The incident light is incident on the surface of the diffraction grating at a 30-degree angle. Light of different wavelengths exits at different angles after diffraction, and the angle difference is converted into a spatial position difference by a collimating lens. After wavelength unfolding, the reflected spectrum forms a one-dimensional distribution in space, with shorter wavelengths distributed on one side and longer wavelengths distributed on the other side. The spatial position of different wavelengths of light has a linear correspondence with their wavelength values.

[0027] refer to Figure 1 The entire monitoring system is built upon a sophisticated optical sensing link. Its core logic lies in directly compressing and acquiring high-dimensional spectral information using optical hardware. The system's optical path begins with the broadband light source on the left side of the diagram. In a preferred embodiment, this light source is an amplified spontaneous emission source, configured to output a center wavelength... And 3 dB bandwidth Continuous broadband light, with output optical power maintained at To ensure a high signal-to-noise ratio for long-distance transmission, a broadband beam is coupled into port 1 of the fiber optic circulator via a single-mode fiber and then guided to the monitoring site with low loss through port 2. The colored-filled area at the bottom of the diagram represents the sensing region extending into the tunnel boring machine (TBM). This section, with its ultra-weak grating array sensing fiber arranged along the inner wall of the tunnel segment, serves as the sensory nerve endings of the entire system. This sensing fiber is composed of a massive number of ultra-weak fiber Bragg gratings continuously etched into the core of a single-mode fiber, with the physical spacing between adjacent gratings set to [value missing]. And the reflectivity of each grating Strictly controlled Within an extremely low range, this design effectively suppresses crosstalk and cumulative loss of optical signals during transmission, enabling a single optical fiber to accommodate thousands of sensing points. When broadband light propagates in the sensing fiber, each ultra-weak fiber Bragg grating selectively reflects a specific wavelength signal that matches its local refractive index and grating period. These reflected light signals are transmitted in reverse and converge to form a composite reflection spectrum containing all distributed information.

[0028] After the composite reflection spectrum is relayed back to the port of the fiber optic circulator, it is output from the port and enters the spectral analysis subsystem. The beam is first incident on the dispersive element; in this embodiment, a line density of [insert line density here] is selected. A transmission diffraction grating. Through the diffraction of the dispersive element, the spectral signal, originally mixed in the time and frequency domains, is converted into a distributed signal in the spatial domain. This crucial physical process is clearly depicted in the figure by two dashed lines of specific colors: the red dashed line represents the propagation path of the long-wavelength light component, and the blue dashed line represents the propagation path of the short-wavelength light component. The fan-shaped area formed by these two colored dashed lines visually reveals the linear mapping relationship between wavelength and spatial position; that is, light of different wavelengths is diffracted to different spatial angles, and after being transformed by subsequent collimating lenses, is projected onto different lateral positions on the plane of the spatial light modulator. The spatial light modulator, located at the end of the optical path, employs a digital micromirror array, which includes... The system comprises tiny mirror units arranged along the spectral spread direction. Each micromirror's deflection state is independently controlled by loading a Bernoulli random binary pattern. Micromirrors in the on state reflect and converge light of the corresponding wavelength to a single-pixel photodetector, while those in the off state deflect light to the absorption end. The single-pixel photodetector integrates the received light energy and outputs an analog voltage value, which is then converted from analog to digital to form a compressed sampling vector. One of the elements. This process is mathematically equivalent to calculating the original spectral vector. row vectors of the currently loaded random measurement matrix The inner product between them. By rapidly switching random patterns on the digital micromirror array and simultaneously recording detector readings, the system can acquire compressed data containing complete wavelength distribution information in far fewer scans than required by the Nyquist sampling theorem, providing the necessary hardware observation foundation for subsequent sparse reconstruction algorithms.

[0029] The wavelength-expanded optical signal is projected onto a spatial light modulator for modulation. A spatial light modulator is an optical device capable of spatially modulating an incident light field; in this embodiment, a digital micromirror array (DMI) is used. The DMI consists of a large number of micromirrors arranged in a two-dimensional array. Each micromirror is approximately 10 micrometers by 10 micrometers in size, and its deflection angle can be independently controlled. When a micromirror is in the open state, its reflective surface faces the photodetector, and light incident on the micromirror is reflected to the photodetector. When a micromirror is in the closed state, its reflective surface is deflected at a certain angle, and the incident light is deflected to the absorption end and does not enter the photodetector. By controlling the on / off state of each micromirror in the DMI, selectively allowing certain wavelengths of light to enter the photodetector can be achieved, thus realizing spatial modulation of the reflected spectrum.

[0030] The random modulation pattern loaded on the digital micromirror array adopts the form of a Bernoulli random binary pattern. The Bernoulli random binary pattern is generated as follows: for each micromirror element in the digital micromirror array, it is independently and randomly set to an on or off state with a probability of 0.5, and the states of each micromirror element are independent of each other. The Bernoulli distribution is chosen as the distribution form of the random modulation pattern based on the requirements of compressed sensing theory. The Bernoulli random matrix satisfies the restricted isometry condition in compressed sensing theory, which can guarantee the convergence and reconstruction accuracy of the subsequent compressed sensing reconstruction algorithm. In a specific implementation, the digital micromirror array contains 1024 micromirror elements arranged along the wavelength expansion direction, each micromirror element corresponding to a wavelength range of approximately 0.04 nanometers in the reflection spectrum.

[0031] The process of acquiring the compressed sampling vector is as follows: First, a first Bernoulli random binary pattern is loaded onto the digital micromirror array. Based on this pattern, the on / off states of each micromirror element are set. Micromirror elements in the on state reflect light of the corresponding wavelength to the photodetector, while micromirror elements in the off state deflect light of the corresponding wavelength to the absorption end. The photodetector integrates all the received light and records the integrated light intensity value. Then, a second Bernoulli random binary pattern is used, and the above process is repeated to record the second integrated light intensity value. Several different Bernoulli random binary patterns are loaded sequentially, and the integrated light intensity value is recorded after each pattern is loaded. Finally, all the integrated light intensity values ​​are arranged in the loading order to form the compressed sampling vector.

[0032] The length of the compressed sampling vector is much smaller than the number of sampling points in the original reflection spectrum, which is the core advantage of compressed sensing technology. Traditional spectral sampling methods require sampling each wavelength point individually, with the number of sampling points equal to the wavelength resolution points. Compressed sensing, however, uses random modulation to compress and encode the original spectral information into a small number of measurements, significantly reducing the number of sampling points. According to compressed sensing theory, when the original signal exhibits sparsity in a certain transform domain, only a small number of random measurements are needed to losslessly recover the original signal. For the reflection spectrum of ultra-weak fiber grating array sensing fibers, it exhibits good sparsity in the discrete cosine transform or Fourier transform domain because the reflection spectrum only has sharp reflection peaks near the Bragg wavelength positions of each ultra-weak fiber Bragg grating, with reflection intensity close to zero at other wavelength positions. In one specific implementation, the original reflection spectrum contains 5000 wavelength sampling points, while the compressed sampling vector contains only 800 measurements, achieving a compression ratio of 6.25 times. This significantly reduces data acquisition time and storage requirements.

[0033] The compressed sampling process can be described mathematically as follows: ,in Represents the compressed sampling vector. This represents the observation matrix consisting of all Bernoulli random binary patterns arranged in rows. This represents the wavelength distribution vector of the original reflectance spectrum. Observation matrix. The number of rows is equal to the length of the compressed sampling vector, the number of columns is equal to the number of wavelength sampling points of the original reflection spectrum, and the elements in the matrix take values ​​of 0 or 1, corresponding to the off and on states of the micromirror element, respectively.

[0034] After obtaining the compressed sampling vector, the original reflection spectrum wavelength distribution needs to be recovered using a compressed sensing reconstruction algorithm, and then the center wavelength of each ultra-weak fiber Bragg grating needs to be calculated. Considering the structural characteristics of shield tunnels, this embodiment proposes a segmented cascaded compressed sensing reconstruction method based on segment ring constraints.

[0035] A shield tunnel is constructed by splicing several prefabricated segment rings along the longitudinal direction of the tunnel. Each segment ring is a relatively independent structural unit, and the segments within a ring are connected by bolts to form a closed loop. Adjacent segment rings are also connected by bolts. This structural characteristic determines that tunnel deformation has a segmented and continuous feature: within a single segment ring, structural deformation exhibits good continuity and correlation; however, at the boundaries of adjacent segment rings, due to the presence of ring joints, deformation may show a certain degree of discontinuity. The segmented cascaded reconstruction method in this embodiment utilizes this structural feature, segmenting the ultra-weak fiber Bragg array sensing fiber according to the segment ring division. Each sensing segment corresponds to a segment ring and includes all ultra-weak fiber Bragg gratings within the range of that segment ring.

[0036] In one specific implementation, the width of the tunnel segment ring is 1.5 meters, and the spacing of the ultra-weak fiber Bragg gratings is 0.5 meters. Therefore, each segment ring contains 3 ultra-weak fiber Bragg gratings, and each sensing segment corresponds to 3 gratings. For a tunnel with a length of 2.5 kilometers, it is divided into approximately 1667 sensing segments.

[0037] The first step in segmented compressed sensing reconstruction is to construct the local observation relationships for each sensing segment. Since compressed sampling is performed simultaneously on all ultra-weak fiber Bragg gratings, the compressed sampling vector contains the wavelength information of all gratings. In order to reconstruct a specific sensing segment individually, it is necessary to extract the local observation matrix related to that sensing segment from the overall observation matrix.

[0038] The specific extraction method is as follows: First, determine the spatial position of the Bragg wavelength of each ultra-weak fiber Bragg grating within the current sensing segment after wavelength expansion by the dispersive element; that is, the range of micromirror element numbers covered by these wavelength values ​​on the digital micromirror array. Then, extract the switching state values ​​at these micromirror element number positions from each Bernoulli random binary pattern, arrange them into column vectors according to the pattern loading order, and then horizontally combine the column vectors to form the local observation matrix of the current sensing segment. The number of rows in the local observation matrix is ​​equal to the total number of loaded Bernoulli random binary patterns, and the number of columns is equal to the number of micromirror elements covered by the ultra-weak fiber Bragg grating within the current sensing segment after wavelength expansion.

[0039] Simultaneously, it is necessary to separate the local compressed sampling sub-vectors corresponding to the current sensing segment from the compressed sampling vector. Since each measurement value in the compressed sampling vector is the light intensity integral across all wavelengths, it theoretically contains the information contributions of all sensing segments and cannot be directly separated. This embodiment adopts an approximate separation method: assuming that the reflection peaks of each ultra-weak fiber Bragg grating do not overlap in the wavelength domain, it can be considered that the contributions from different sensing segments in each measurement value are additive, and the contributions of each sensing segment are gradually separated through iterative reconstruction. In the initial stage, the compressed sampling vector is directly used as the initial estimate of the local compressed sampling sub-vectors of each sensing segment.

[0040] The core algorithm of segmented compressed sensing reconstruction adopts the iterative hard thresholding reconstruction method. Iterative hard thresholding reconstruction is a computationally efficient compressed sensing reconstruction algorithm that gradually approximates the original signal by alternately executing two steps: gradient descent and hard thresholding.

[0041] For each sensing segment, the execution process of iterative hard threshold reconstruction is as follows: First, the center wavelengths of each ultra-weak fiber Bragg grating within the current sensing segment are initialized, with the initial values ​​set to the design nominal wavelength of each grating. The design nominal wavelength is a theoretical wavelength value preset during grating inscription; under strain-free and reference temperature conditions, the actual Bragg wavelength of the grating should be equal to the design nominal wavelength. The initialized wavelength estimates are then combined to form a wavelength estimation vector. .

[0042] A sparse representation basis is chosen for the sparse transformation of the signal. The role of the sparse representation basis is to transform the original wavelength-shifted signal to another domain, where the signal exhibits sparsity, meaning most transform coefficients are close to zero, with only a few coefficients having large amplitudes. In this embodiment, the Discrete Cosine Transform (DCT) is used as the sparse representation basis in the first round of reconstruction. The DCT can decompose a signal in the spatial domain into a superposition of different frequency components. For the strain distribution signal along the longitudinal direction of a shield tunnel, its main energy is concentrated in the low-frequency components because tunnel structural deformation is usually slow and does not exhibit violent spatial oscillations. Choosing the DCT basis ensures that the strain signal exhibits good sparsity in the transform domain.

[0043] Calculate the transform coefficients of the current wavelength estimate under the sparse representation basis. ,in Represents the discrete cosine transform matrix. Indicates the first Wavelength estimation vector for the next iteration This represents the transformation coefficient vector.

[0044] A hard thresholding operation is performed on the transform coefficient vector, retaining the top few coefficients with the largest absolute values ​​and forcing the remaining coefficients to zero. The mathematical form of hard thresholding is: for each transform coefficient, if its absolute value ranks first... If the bit is not specified, its original value is retained; otherwise, it is set to zero. Here This is a preset sparsity parameter, representing the number of non-zero coefficients of the signal in the transform domain. In a specific implementation, the sparsity parameter... The threshold is set to twice the number of gratings within the sensing segment. The hard threshold truncation operation forces the reconstruction result to be sparsity, filtering out noise and interference components.

[0045] refer to Figure 4 The horizontal axis in the figure represents the index of the discrete cosine transform coefficients. The range is from 1 to 60, and the vertical axis represents the absolute value of the coefficient. The range is 0 to 100. The blue bar chart shows the amplitude distribution of each DCT coefficient. The chart clearly shows that the signal exhibits good sparsity characteristics in the transform domain: low-frequency coefficients (those with smaller numbers) have larger amplitudes, with the amplitudes of coefficients 1 to 5 being approximately 85, 72, 58, 35, and 28 respectively. These low-frequency components carry the main information about the wavelength distribution; while high-frequency coefficients (those with larger numbers) have amplitudes close to zero or contain only a small amount of noise. This sparsity distribution characteristic is consistent with the physical characteristics of shield tunnel structural deformation, because the strain distribution along the longitudinal direction of the tunnel is usually a slowly changing continuous function, without violent spatial oscillations; therefore, its spectral energy is mainly concentrated in the low-frequency range. The red horizontal dashed line represents the hard threshold cutoff line. The value is set to 15. The hard thresholding operation retains coefficients whose absolute value exceeds the threshold and forces the remaining coefficients to zero. In the graph, dark blue bars represent the retained coefficients, and light blue bars represent the coefficients set to zero. The blue semi-transparent area indicates the region of retained coefficients, and the gray semi-transparent area indicates the region of zeroed coefficients. The sparsity is indicated in the upper right corner. A value of 7 indicates that the top 7 largest coefficients are retained. The mathematical expression for hard thresholding is: for the transform coefficient vector... , retain satisfaction The coefficients are calculated by setting the coefficients that do not meet the conditions to zero, thus obtaining the truncated coefficient vector. This operation forces the reconstruction results to be sparse, which can effectively filter out measurement noise and random interference components, and improve the accuracy and stability of wavelength reconstruction.

[0046] Perform an inverse discrete cosine transform on the retained transform coefficients to obtain the updated wavelength estimate. ,in This represents the vector of transform coefficients after hard thresholding.

[0047] Calculate the reconstruction residual, which is the difference between the product of the local observation matrix and the updated wavelength estimate and the local compressed sample subvector. ,in Represents a locally compressed sampled subvector. Represents the local observation matrix. This represents the residual vector. The residual vector reflects the deviation between the current wavelength estimate and the true value; the smaller the residual, the closer the reconstruction is to the true signal.

[0048] The residual vector is propagated backward along the transpose of the local observation matrix and superimposed onto the current wavelength estimate to form the corrected wavelength estimate. ,in This represents the step size parameter, which controls the correction magnitude in each iteration. The selection of the step size parameter needs to balance convergence speed and stability; an excessively large step size may cause iteration divergence, while an excessively small step size will result in slow convergence. In a specific implementation, the step size parameter... Set as ,in This represents the spectral norm of the local observation matrix.

[0049] Repeat the process of calculating transform coefficients, hard thresholding, inverse transform, residual calculation, and backpropagation correction until the L2 norm of the residual vector is reached. The residual is less than a preset residual convergence threshold. In one specific implementation, the residual convergence threshold is set to 1% of the initial residual. When the convergence condition is met, the first reconstructed wavelength sequence of the current sensing segment is output.

[0050] After completing the initial independent reconstruction of all sensor segments, boundary continuity cascade correction is required. The initial reconstruction is performed independently for each sensor segment, without considering the continuity constraints between adjacent segments. Therefore, discontinuous jumps in wavelength values ​​may occur at the boundaries of sensor segments. These jumps are due to the inherent accumulation of errors from the independent segment reconstruction and do not reflect the true state of the tunnel structure. The purpose of boundary continuity correction is to eliminate this artificial discontinuity and restore the continuity of the wavelength distribution curve across the entire tunnel.

[0051] The specific execution process of boundary continuity correction is as follows: At the junction of two adjacent sensing segments, extract several wavelength values ​​located at the end of the first-round reconstructed wavelength sequence of the preceding sensing segment on the boundary side, and simultaneously extract several wavelength values ​​located at the beginning of the first-round reconstructed wavelength sequence of the following sensing segment on the boundary side. In a specific implementation, two wavelength values ​​are extracted from each side of the boundary.

[0052] refer to Figure 2In this embodiment, the total tunnel length is 2500 meters. Ultra-weak fiber Bragg grating array sensing fibers are continuously deployed along the tunnel's longitudinal direction. The spacing between the ultra-weak fiber Bragg gratings etched on the fibers is 0.5 meters, with a total of 5000 grating measurement points covering the entire tunnel. In the figure, the horizontal axis represents the tunnel mileage in meters, and the vertical axis represents the Bragg wavelength of each grating in nanometers. The blue solid curve shows the wavelength distribution result obtained after processing by the segmented cascaded compressed sensing reconstruction algorithm. The reference wavelength of this curve is 1550 nanometers, and the wavelength offset range is approximately ±0.3 nanometers. The curve exhibits obvious spatial variation characteristics, containing multiple local peaks and troughs. These fluctuations reflect the combined changes in strain and temperature along the tunnel. Near 800 meters and 1600 meters, the wavelength curve shows a significant local decrease, due to leakage in this area leading to a temperature drop and consequently a negative wavelength shift in the gratings. The red inverted triangles in the figure indicate the locations of the temperature reference gratings. These gratings are encapsulated using loose tubes, which provide mechanical isolation and protection, preventing mechanical coupling between the gratings and the tunnel structure, thus eliminating strain transfer effects. The temperature reference gratings respond only to temperature changes, and their wavelength shift can be directly used to calculate the temperature value at that location. In this embodiment, the temperature reference gratings are spaced 50 meters apart, with a total of 51 gratings deployed along the 2500-meter tunnel. The parameter description box in the upper right corner of the figure indicates the 50-meter spacing of the temperature reference gratings and the technical characteristics of the loose tubes in eliminating strain transfer. By spaced temperature reference gratings along the sensing fiber path, discrete temperature measurement points can be obtained. Combined with linear interpolation methods, a continuous temperature distribution curve can be reconstructed, providing fundamental data for subsequent strain and temperature separation calculations.

[0053] Based on the extracted wavelength values ​​at the end, linear extrapolation is performed to predict the wavelength values ​​at the boundary location. Linear extrapolation is based on the assumption that the wavelength values ​​at the end change linearly. The slope and intercept of the linear change are determined through least-squares fitting, and then extrapolated to the boundary location. Similarly, linear back extrapolation is performed based on several wavelength values ​​at the beginning to predict the wavelength values ​​at the boundary location.

[0054] The difference between the linear extrapolation prediction value of the previous sensing segment and the wavelength value at the beginning of the next sensing segment is calculated to obtain the boundary jump variable. At the same time, the difference between the linear backward extrapolation prediction value of the next sensing segment and the wavelength value at the end of the previous sensing segment is calculated to obtain the backward boundary jump variable.

[0055] Half of the boundary jump variable is used as the boundary correction amount for the preceding sensing segment, and half of the reverse boundary jump variable is used as the boundary correction amount for the following sensing segment. This 50 / 50 allocation method is based on the principle of symmetry, assuming that the error of the boundary jump is contributed equally by the sensing segments on both sides.

[0056] When applying boundary corrections to the reconstructed wavelength sequence of the previous sensing segment, a linear decreasing approach is used instead of applying the same correction to all wavelength values. The wavelength value closest to the boundary receives the full boundary correction, the wavelength value farthest from the boundary receives zero correction, and the correction for wavelength values ​​in the middle is determined by linear interpolation proportional to their distance from the boundary. This linear decreasing approach ensures a smooth transition of the corrected wavelength sequence within the sensing segment, avoiding the introduction of new discontinuities. The same linear decreasing approach is used to apply boundary corrections to the subsequent sensing segment.

[0057] The boundary continuity correction described above is performed at the junctions of all adjacent sensing segments to obtain the corrected wavelength sequences of each sensing segment after boundary correction.

[0058] After completing the boundary continuity correction, a second-round iterative refinement reconstruction is performed to further improve the reconstruction accuracy. The second-round reconstruction uses the reconstructed wavelength sequences of each sensing segment after correction as new initial values ​​and re-executes the iterative hard-threshold reconstruction process. Unlike the first-round reconstruction, the second-round reconstruction replaces the sparse representation basis with a local Fourier basis constructed centered on the corrected reconstructed wavelength sequences.

[0059] The considerations for using a local Fourier basis are as follows: The discrete cosine transform basis used in the first-round reconstruction is a global basis, suitable for coarse estimation of the signal. Secondary refined reconstruction requires fine adjustments based on the first-round results, where the wavelength change is relatively small. The local Fourier basis is a transform basis constructed within a limited frequency range, centered on the first-round reconstruction results. The frequency range of its basis vectors is limited to the typical frequency range of wavelength changes caused by shield tunnel structural deformation and leakage. This frequency range is determined based on the spatial characteristic scale of tunnel structural deformation, which typically ranges from several meters to tens of meters in spatial wavelength, corresponding to a frequency range of approximately 0.02 to 0.5 periods per meter. Using a local Fourier basis allows the secondary reconstruction to focus more on the wavelength shift caused by actual physical changes, filtering out high-frequency noise and low-frequency drift unrelated to structural deformation.

[0060] After completing the second iteration of refined reconstruction, the refined reconstruction wavelength sequences of all sensing segments are spliced ​​together in spatial order to form a reconstructed wavelength distribution curve distributed along the entire length of the shield tunnel. The reconstructed wavelength distribution curve describes the current Bragg wavelength value of each ultra-weak fiber Bragg grating on the ultra-weak grating array sensing fiber.

[0061] Obtaining strain and temperature distribution curves from the reconstructed wavelength distribution curve is a classic problem in fiber optic sensing. The Bragg wavelength of a fiber Bragg grating is sensitive to both strain and temperature; when strain or temperature changes occur at the grating's location, the Bragg wavelength shifts. To obtain independent strain and temperature distributions, this embodiment employs a temperature reference grating method for separate strain and temperature calculations.

[0062] refer to Figure 3 In the graph, the horizontal axis represents the tunnel mileage, ranging from 0 to 2500 meters, and the vertical axis represents the temperature value in degrees Celsius, ranging from 17 to 26 degrees Celsius. The red circular scatter dots represent the measured temperature values ​​of each temperature reference grating, calculated based on the wavelength offset of the temperature reference grating and a pre-calibrated temperature sensitivity coefficient. Temperature sensitivity coefficient. This represents the wavelength shift caused by a unit change in temperature. For a standard quartz fiber Bragg grating, this coefficient is approximately 10 picometers per degree Celsius. The temperature calculation formula is as follows: ,in For reference temperature, This represents the wavelength offset. The solid red line curve represents the continuous temperature distribution obtained through linear interpolation. This curve connects the measurement points of adjacent temperature reference gratings and performs linear interpolation in the middle section. The dashed blue line curve represents the actual temperature distribution along the tunnel as a reference. The gray horizontal dotted line indicates a reference temperature of 22 degrees Celsius. Near 800 meters and 1600 meters, the temperature curve shows a significant local decrease, with a drop of approximately 2 to 2.5 degrees Celsius. These two areas of temperature drop caused by leakage are marked with semi-transparent blue areas in the figure. The arrows illustrate the physical mechanism of the temperature drop: the groundwater temperature is usually maintained at around 15 to 18 degrees Celsius, close to the local annual average air temperature, lower than the ambient temperature inside the tunnel. When leakage occurs, the low-temperature groundwater enters the tunnel and flows or evaporates near the leakage point, carrying away heat and causing a local temperature drop. This negative temperature bias is an important criterion for leakage detection; potential leakage locations can be identified by monitoring local low-temperature anomalies in the temperature distribution curve.

[0063] Several temperature reference gratings are spaced along the fiber optic cable routing path of the ultra-weak grating array sensing fiber. The temperature reference gratings are loosely packaged, providing mechanical isolation and protection to prevent mechanical coupling between the grating and external structures. This eliminates strain transmission, allowing the temperature reference gratings to respond only to temperature changes and not strain changes. In one specific implementation, a temperature reference grating is placed every 50 meters; for a 2.5-kilometer-long tunnel, a total of 50 temperature reference gratings are used.

[0064] The wavelength values ​​at each temperature reference grating position are extracted from the reconstructed wavelength distribution curve. Since the temperature reference grating only responds to temperature changes, its wavelength offset can be directly used to calculate the temperature value. The temperature value at each temperature reference grating position is obtained by dividing the wavelength offset relative to the nominal wavelength by a pre-calibrated temperature sensitivity coefficient. The temperature sensitivity coefficient refers to the wavelength offset caused by a unit temperature change; for a standard quartz fiber Bragg grating, the temperature sensitivity coefficient is approximately 10 picometers per degree Celsius.

[0065] Linear interpolation is used to obtain a continuous temperature distribution curve for the sections between adjacent temperature reference gratings. Linear interpolation assumes that the temperature varies linearly between adjacent temperature reference gratings, which is a reasonable assumption in tunnel environments because the temperature field inside a tunnel typically changes slowly. Through interpolation, the discrete temperature reference grating measurement points are expanded into a continuous temperature distribution curve covering the entire tunnel length.

[0066] After obtaining the temperature distribution curve, the wavelength shift component caused by temperature change can be calculated. Subtracting the wavelength shift component caused by temperature, as determined by the temperature distribution curve, from the wavelength shift at each position in the reconstructed wavelength distribution curve yields the wavelength shift caused by pure strain. Dividing the wavelength shift caused by pure strain by a pre-calibrated strain sensitivity coefficient converts it into a strain distribution curve. The strain sensitivity coefficient refers to the wavelength shift caused by a unit strain; for a standard silica fiber Bragg grating, the strain sensitivity coefficient is approximately 1.2 picometers per microstrain.

[0067] The obtained strain and temperature distribution curves describe the strain and temperature states at various locations along the longitudinal direction of the shield tunnel, providing fundamental data for subsequent joint determination of deformation and leakage. The strain distribution curve reflects the structural deformation of the tunnel segments; significant tensile or compressive strain at a location indicates possible structural deformation. The temperature distribution curve helps determine the occurrence of leakage because groundwater temperature is typically lower than the ambient temperature inside the tunnel. When leakage occurs, the leaking water carries away heat, causing a local temperature drop, which is reflected as a negative local temperature deviation on the temperature distribution curve.

[0068] After obtaining the strain and temperature distribution curves along the shield tunnel, these curves need to be analyzed and processed to extract characteristic indicators that can characterize the abnormal structural state. Based on this, the anomaly types of each tunnel section are determined and marked. The abnormal states that may occur during the operation of a shield tunnel mainly include two categories: structural deformation and leakage. These two types of anomalies are often intrinsically related: structural deformation may cause the joints of the tunnel segments to open, leading to leakage, while continuous leakage will exacerbate the loss of surrounding soil and further worsen the structural deformation. Therefore, this embodiment adopts a joint determination method, considering both strain and temperature anomalies simultaneously, to identify simple deformation anomalies, simple leakage anomalies, and composite anomalies that are coupled with both.

[0069] To achieve a systematic scanning analysis of the entire tunnel length, this embodiment employs a sliding window method to traverse the strain and temperature distribution curves. The sliding window is a classic signal analysis method that effectively captures local variations in a signal by moving a fixed-length analysis window across the signal sequence and calculating local characteristic indices segment by segment. The sliding window is chosen instead of performing a global analysis of the entire curve because tunnel anomalies typically manifest as local features, and a global analysis would overwhelm these local anomaly signals with a large amount of normal data in the background.

[0070] The length of the sliding window needs to be set by comprehensively considering the tunnel's structural characteristics and anomaly detection sensitivity. A window length that is too small can lead to unstable statistical features and false alarms; a window length that is too large may merge multiple adjacent independent anomaly segments into one, reducing spatial positioning accuracy. The segment rings of a shield tunnel are the basic structural units, and anomalies often occur within a single segment ring or within a few adjacent segment rings. In one specific implementation, the sliding window length is set to 3 meters, corresponding to the width of two segment rings. This length ensures that the window contains enough sampling points to obtain stable statistical features while maintaining good spatial resolution. The sliding window moves longitudinally along the tunnel with a fixed step size, typically set to half the window length, i.e., 1.5 meters. This results in a 50% overlap between adjacent windows, preventing anomaly segments located precisely at the window boundary from being missed.

[0071] For the strain distribution curve, this embodiment extracts the strain amplitude index as a characteristic quantity characterizing structural deformation. The strain amplitude index is defined as the difference between the maximum and minimum strain values ​​within a sliding window, mathematically expressed as: ,in This indicates the strain amplitude index of the current sliding window. This indicates the maximum strain value of the strain distribution curve within the current sliding window. This represents the minimum strain value of the strain distribution curve within the current sliding window. The strain amplitude, rather than the average strain or absolute strain value, is chosen as the characteristic index based on the following considerations: When a tunnel structure undergoes local deformation, a complex strain distribution of tension and compression occurs within the deformation area. The convex side of the bending deformation is under tension, while the concave side is under compression, thus significantly increasing the strain amplitude in the deformation area. In contrast, the slow settlement of the entire tunnel or the uniform thermal strain caused by temperature will cause the strain value to rise or fall synchronously throughout the entire section, without significantly altering the strain amplitude within a local area. Using the strain amplitude index can effectively distinguish between local structural deformation and overall uniform changes, improving the targeting of deformation detection.

[0072] For the temperature distribution curve, this embodiment extracts the negative temperature deviation index as a characteristic quantity to characterize leakage. Leakage detection utilizes the temperature difference between groundwater and the air inside the tunnel. The temperature of groundwater is usually maintained near the local annual average air temperature, approximately 15 to 18 degrees Celsius in most areas, while the temperature inside the tunnel is usually higher than the groundwater temperature due to the heat generated by train operation and the effect of the ventilation system. When leakage occurs, the low-temperature groundwater enters the tunnel and evaporates or flows near the leakage point, carrying away heat and causing the temperature in the area surrounding the leakage point to drop. This local temperature drop is represented by a negative deviation on the temperature distribution curve, i.e., the temperature is lower than the temperature level of the surrounding normal section.

[0073] The negative temperature bias index is calculated as follows: First, calculate the arithmetic mean of the temperature values ​​in the preceding and following adjacent sliding windows of the current sliding window. Then, average these two temperature averages again to obtain the reference temperature value. Next, calculate the decrease in the average temperature within the current sliding window relative to the reference temperature value. Mathematically, this is expressed as follows: ,in This indicates the negative temperature bias of the current sliding window. This represents the reference temperature value calculated from adjacent sliding windows. This represents the average temperature of the temperature distribution curve within the current sliding window. A positive value indicates that the current window temperature is lower than the surrounding area, which may indicate leakage; when... A negative value or value close to zero indicates that the current window temperature is normal or slightly high, and the possibility of leakage is low. Using the average temperature of adjacent windows as a reference, rather than a fixed absolute temperature threshold, is to accommodate natural temperature variations along the tunnel route. Due to differences in burial depth, geological conditions, and ventilation, background temperatures may vary significantly in different sections of the tunnel. Using a local relative comparison method can eliminate the influence of these background differences and improve the robustness of leakage detection.

[0074] After calculating the feature indicators for all sliding windows, each sliding window is classified and marked according to a preset anomaly threshold. Setting the anomaly threshold is a key parameter for balancing detection sensitivity and false alarm rate. A threshold that is too low will cause many normal fluctuations to be misjudged as anomalies, while a threshold that is too high may miss genuine abnormal events. This embodiment sets strain anomaly thresholds and temperature anomaly thresholds respectively.

[0075] The strain anomaly threshold is determined based on the allowable strain of the tunnel structure and the material strength. The strain level of concrete segments under normal service conditions is typically within ±200 microstrains. When the local strain amplitude exceeds a certain limit, it indicates that the structure may have undergone deformation beyond the normal range. In one specific implementation, the strain anomaly threshold is set at 150 microstrains; that is, when the strain amplitude index of a certain sliding window exceeds 150 microstrains, a strain anomaly is considered to exist within that window. This threshold setting considers the measurement accuracy of the fiber optic sensing system and the normal strain fluctuation range of the tunnel structure, enabling the detection of engineering-significant structural deformations without generating excessive false alarms due to measurement noise.

[0076] The temperature anomaly threshold is determined based on the typical temperature change range caused by leakage. According to field monitoring experience, when significant leakage occurs, the temperature drop near the leakage point is typically between 0.5 and 3 degrees Celsius, with the specific value depending on factors such as the leakage volume, groundwater temperature, and ambient temperature differences. In one specific implementation, the temperature anomaly threshold is set at 0.8 degrees Celsius; that is, when the negative temperature deviation index of a sliding window exceeds 0.8 degrees Celsius, the window is considered to have a temperature anomaly. This threshold can detect leakage events with a certain amount of leakage, while minor temperature changes caused by non-leakage factors such as minute amounts of seepage or condensation will not trigger an alarm.

[0077] Based on the comparison results of strain amplitude and temperature negative deviation indices with their respective thresholds, each sliding window is marked with anomaly type. The marking rules are as follows: When the strain amplitude index of a sliding window exceeds the strain anomaly threshold but the temperature negative deviation index does not exceed the temperature anomaly threshold, the sliding window is marked as a deformation anomaly window, indicating that there is structural deformation in this section but no signs of leakage are detected; when the temperature negative deviation index of a sliding window exceeds the temperature anomaly threshold but the strain amplitude index does not exceed the strain anomaly threshold, the sliding window is marked as a suspected leakage window, indicating that there are signs of leakage in this section but the structural deformation is within the normal range; when the strain amplitude index of a sliding window exceeds the strain anomaly threshold and the temperature negative deviation index also exceeds the temperature anomaly threshold, the sliding window is marked as a leakage-deformation coupling window, indicating that there are both structural deformation and leakage phenomena in this section, and the two may have an interactive coupling relationship. The leakage-deformation coupling window represents the most serious anomaly state, because the simultaneous existence of deformation and leakage often means that the structural damage has reached a high level and needs to be addressed first.

[0078] In an optional implementation, in addition to the aforementioned binary threshold determination method, a multi-level threshold determination method can be used to obtain a more refined classification of the degree of anomaly. Specifically, two levels are set for the strain amplitude index: a mild anomaly threshold and a severe anomaly threshold, such as 100 microstrain and 200 microstrain; similarly, two levels are set for the temperature negative deviation index: a mild anomaly threshold and a severe anomaly threshold, such as 0.5 degrees Celsius and 1.5 degrees Celsius. Based on the combination of threshold intervals into which the characteristic index falls, the abnormal state can be further subdivided into various types such as mild deformation, severe deformation, mild leakage, severe leakage, mild coupling, and severe coupling, providing richer information for subsequent graded alarms and handling decisions.

[0079] In another alternative implementation, time-dimensional analysis can be introduced to improve the reliability of anomaly detection. Specifically, not only are the strain and temperature distribution curves at the current moment analyzed, but also the distribution curves from the previous several sampling periods are combined for time-series analysis. An anomaly window is only confirmed as such when it is determined to be abnormal in multiple consecutive sampling periods; situations where an anomaly occurs only in a single sampling period and subsequently returns to normal are considered transient disturbances and no anomaly marker is generated. This time-series filtering method can effectively suppress false alarms caused by measurement noise or transient interference, improving the accuracy of anomaly detection.

[0080] After marking all sliding windows for anomalies, the system needs to generate alarm information and plan inspection routes based on the marking results to guide maintenance personnel in timely detection and handling of tunnel anomalies. The generation of alarm information and the planning of inspection routes require comprehensive consideration of the severity and spatial distribution of the anomalies to ensure that limited maintenance resources are prioritized for the most urgent areas.

[0081] The alarm levels are classified according to the severity of the anomaly type. Different types of anomalies pose different threats to the safety of the tunnel structure, and therefore require different response measures and response times. This embodiment classifies alarm levels into three levels, from highest to lowest: Level 1 alarm, Level 2 alarm, and Level 3 alarm.

[0082] A leakage-deformation coupling window generates a Level 1 alarm. The leakage-deformation coupling state represents the most severe anomaly, as the simultaneous existence of structural deformation and leakage often creates a vicious cycle: structural deformation causes the segment joints to open or crack, providing a channel for groundwater infiltration; the infiltrated groundwater then erodes and removes the infill material and soil around the joints or cracks, weakening the structure's support and exacerbating the deformation. If this coupling effect is not addressed promptly, it can rapidly deteriorate within a short period, threatening the tunnel structure's safety. Therefore, the leakage-deformation coupling window is assigned the highest Level 1 alarm, requiring maintenance personnel to arrive on-site for inspection and handling as quickly as possible.

[0083] A suspected leakage window generates a Level 2 alarm. While simple leakage does not directly threaten the structural load-bearing capacity, continuous leakage can cause multiple adverse effects: leaking water may corrode metal facilities and electrical equipment within the tunnel, affecting operational safety; accumulated leaking water within the tunnel may affect train operation; and long-term leakage can gradually erode surrounding soil, potentially inducing subsequent structural deformation. Furthermore, the suspected leakage window is marked based on a negative temperature bias index. Although a temperature drop is a typical characteristic of leakage, it can also be caused by other factors, such as abnormal local ventilation or changes in equipment heat dissipation. Therefore, it is called suspected leakage rather than confirmed leakage and requires on-site verification. The suspected leakage window is assigned a Level 2 alarm level, requiring maintenance personnel to arrange an inspection within a specified timeframe.

[0084] An abnormal deformation window generates a Level 3 alarm. Simple structural deformation typically develops slowly in its initial stages, and as long as the deformation does not exceed the structural allowable value, the immediate threat to tunnel safety is relatively low. The presence of an abnormal deformation window indicates that the structural condition of this section warrants attention, and its development trend needs continuous monitoring. If the deformation continues to increase and exceeds a certain limit, reinforcement measures may be necessary. The abnormal deformation window is assigned a Level 3 alarm level and is included in the daily inspection scope for focused monitoring.

[0085] Each alarm message includes the alarm level, an anomaly type label, and a spatial location range. The anomaly type label directly uses the sliding window's marking results and is divided into three types: deformation anomaly, suspected leakage, and leakage-deformation coupling. The spatial location range records the tunnel mileage range corresponding to the sliding window. For example, from kilometer marker K3+150 to K3+153, it indicates that the alarm corresponds to the section from tunnel mileage 3.150 km to 3.153 km. The spatial location information enables maintenance personnel to accurately locate the abnormal section, facilitating on-site search and handling.

[0086] Simultaneously with the generation of alarm information, the system generates an inspection and verification route based on the alarm level and spatial location. The inspection and verification route is a path plan that guides maintenance personnel to check each alarm segment in sequence. Its design goal is to minimize unnecessary round trips during the inspection process and improve inspection efficiency while ensuring that high-level alarms are handled first.

[0087] The process of generating the inspection and verification route is as follows: First, all sliding windows that generated alarms are sorted in descending order of alarm level, with Level 1 alarms at the top, Level 2 alarms next, and Level 3 alarms at the bottom. For sliding windows with the same alarm level, they are then sorted a second time according to their spatial location, i.e., the tunnel mileage, so that alarms of the same level are arranged sequentially along the spatial order of the tunnel.

[0088] After sorting, the tunnel mileage locations corresponding to each alarm sliding window are connected sequentially to form an inspection and verification route. The inspection and verification route is output as an ordered list of mileage locations, where each item contains the alarm level, anomaly type label, and mileage location information. Maintenance personnel proceed to each mileage location in the order of the list to conduct on-site inspections and complete the verification of all alarm sections.

[0089] In one specific implementation, assuming that the joint judgment in step three generates the following alarms: the section from K1+200 to K1+203 is a leakage-deformation coupling window, the section from K2+450 to K2+453 is a suspected leakage window, the section from K0+900 to K0+903 is a deformation anomaly window, and the section from K3+100 to K3+103 is a suspected leakage window. After sorting by alarm level, the first-level alarm from K1+200 to K1+203 is ranked first; the second-level alarms from K2+450 to K2+453 and K3+100 to K3+103 are arranged in mileage order; and the third-level alarm from K0+900 to K0+903 is ranked last. The generated inspection and verification route is: station 1 K1+200, station 2 K2+450, station 3 K3+100, station 4 K0+900. The maintenance personnel started from the tunnel entrance, first went to K1+200 to handle the most urgent level 1 alarm, then went to K2+450 and K3+100 to check level 2 alarms, and finally went to K0+900 to check level 3 alarms.

[0090] The generated inspection and verification route is displayed and output through the monitoring terminal. The monitoring terminal can display the inspection route using a graphical interface, marking the location of each alarm on the tunnel longitudinal profile diagram, and using different colors to distinguish different alarm levels, such as red for level 1 alarms, orange for level 2 alarms, and yellow for level 3 alarms. Detailed alarm information and inspection sequence are also displayed in a list format for easy review and execution by maintenance personnel.

[0091] In one alternative implementation, the generation of the inspection and review route can also consider optimizing the starting point and direction of travel. When the initial position of the inspection personnel is known, the system can calculate the shortest path from the initial position to all alarm points and back. Since the tunnel is a linear structure, this problem simplifies to a traveling salesman problem in one-dimensional space, which can be solved using dynamic programming or a greedy algorithm to find the optimal or near-optimal path. Specifically, under the constraint of prioritizing first-level alarms, the access order of second- and third-level alarms is optimized to minimize the total travel distance. This optimization method can significantly reduce inspection time when there are a large number of alarms that are widely distributed.

[0092] In another optional implementation, alarm levels and inspection priorities can be dynamically adjusted based on the duration and trend of alarms. The system records the abnormal indicator values ​​for each sliding window in each monitoring cycle and analyzes their trends. For windows where abnormal indicators are continuously increasing, their inspection priority is increased even if the current alarm level is low; for windows where abnormal indicators remain stable or decrease, their inspection priority can be appropriately reduced. This dynamic priority adjustment mechanism allows inspection resources to be more accurately allocated to sections with rapidly rising risks, improving the effectiveness of preventative maintenance.

[0093] Furthermore, the inspection and verification routes can be linked with the tunnel operation and dispatch system to avoid peak train traffic periods when scheduling inspection time windows, ensuring the safety of inspection personnel. The system determines recommended inspection time windows for each alarm section based on train timetables and alarm levels. Level 1 alarms are scheduled during the nearest operational gap, while Level 2 and 3 alarms can be handled centrally during nighttime maintenance windows. This collaboration with operation and dispatch maximizes inspection efficiency while ensuring safety.

[0094] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these specific embodiments are merely illustrative. Those skilled in the art can omit, substitute, and modify the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, combining the above method steps to perform substantially the same function and achieve substantially the same result according to substantially the same method falls within the scope of the present invention. Therefore, the scope of the present invention is defined only by the appended claims.

Claims

1. A method for combined monitoring of deformation and leakage of a shield tunnel based on distributed optical fiber sensing, characterized in that, Includes the following steps: Step 1: Lay ultra-weak grating array sensing optical fibers along the inner wall of the shield tunnel segments. Use a broadband light source to input detection light into the ultra-weak grating array sensing optical fibers. After the reflected spectrum is wavelength-expanded by a dispersive element, it is projected onto a spatial light modulator. A random modulation pattern is loaded on the spatial light modulator to modulate the wavelength-expanded light signal. The photodetector records the modulated light intensity value to form a compressed sampling vector. Step 2: Divide the ultra-weak grating array sensing fiber into several sensing segments, perform segmented compressed sensing reconstruction on the compressed sampling vector to obtain the reconstructed wavelength sequence of each sensing segment, perform boundary continuity correction on the reconstructed wavelength sequences of adjacent sensing segments, and splice the reconstructed wavelength sequences of each sensing segment after correction to form a reconstructed wavelength distribution curve. Based on the reconstructed wavelength distribution curve, calculate the strain distribution curve and temperature distribution curve distributed along the shield tunnel. Step 3: Extract strain anomaly features and temperature anomaly features from the strain distribution curve and temperature distribution curve respectively, and mark deformation anomaly, leakage anomaly, or leakage-deformation coupling anomaly for each section of the shield tunnel based on the strain anomaly features and temperature anomaly features. Step 4: Generate alarm information of corresponding levels based on the anomaly markers obtained in Step 3, and generate inspection and verification routes based on the level and spatial location of each alarm information.

2. The method of claim 1, wherein, The ultra-weak fiber Bragg array sensing fiber contains several ultra-weak fiber Bragg gratings arranged in series, and the reflectivity of each ultra-weak fiber Bragg grating is set to be between 0.01% and 0.1%.

3. The method of claim 1, wherein, The spatial light modulator is a digital micromirror array, and the random modulation pattern is a Bernoulli random binary pattern. Each micromirror element in the Bernoulli random binary pattern is independently set to an on or off state with a 50% probability. The micromirror element in the on state reflects the incident light to the photodetector, and the micromirror element in the off state deflects the incident light to the absorption end. Several different Bernoulli random binary patterns are loaded sequentially. The photodetector records the light intensity integral value once after each Bernoulli random binary pattern is loaded. All the light intensity integral values ​​are arranged in the loading order to form a compressed sampling vector.

4. The method of claim 1, wherein, In step two, the ultra-weak fiber Bragg array sensing fiber is divided into several sensing segments according to the segment ring structure of the shield tunnel. Each sensing segment corresponds to a segment ring and includes all ultra-weak fiber Bragg gratings within the corresponding segment ring range.

5. The method of claim 4, wherein, In step two, the segmented compressed sensing reconstruction includes: for each sensing segment, extracting column vectors corresponding to the wavelength expansion positions of each ultra-weak fiber Bragg grating within the current sensing segment from the random modulation pattern; arranging the extracted column vectors according to the pattern loading order to form the local observation matrix of the current sensing segment; separating the local compressed sampling sub-vectors corresponding to the current sensing segment from the compressed sampling vectors; and performing iterative hard threshold reconstruction for each sensing segment. The execution process of iterative hard threshold reconstruction is as follows: initializing the center wavelength of each ultra-weak fiber Bragg grating within the current sensing segment to the design nominal wavelength of each ultra-weak fiber Bragg grating; using discrete cosine transform as the sparse representation basis, calculating the current wavelength estimate. The transformation coefficients under the sparse representation basis are calculated, and the top few coefficients with the largest absolute values ​​are retained while the remaining coefficients are set to zero. An inverse discrete cosine transform is performed on the retained coefficients to obtain the updated wavelength estimate. The difference between the product of the local observation matrix and the updated wavelength estimate and the local compressed sampling subvector is calculated as the residual vector. The residual vector is backpropagated along the transpose of the local observation matrix and superimposed on the current wavelength estimate to form the corrected wavelength estimate. The process of calculating transformation coefficients, retaining thresholds, inverse transformation, calculating residuals and backpropagating correction is repeated until the L2 norm of the residual vector is less than the preset residual convergence threshold. The reconstructed wavelength sequence of the current sensing segment is then output.

6. The method according to claim 5, characterized in that, In step two, boundary continuity correction includes: at the boundary between two adjacent sensing segments, extracting several wavelength values ​​located at the end of the reconstructed wavelength sequence of the preceding sensing segment on the boundary side and several wavelength values ​​located at the beginning of the reconstructed wavelength sequence of the following sensing segment on the boundary side; calculating the boundary jump variable between the linear extrapolated predicted values ​​of the end wavelength values ​​and the beginning wavelength values, and calculating the reverse boundary jump variable between the linear backward extrapolated predicted values ​​of the beginning wavelength values ​​and the end wavelength values; and using half of the boundary jump variable as the boundary jump variable for the preceding sensing segment. The boundary correction is calculated by using half of the reverse boundary jump variable as the boundary correction for the next sensing segment. For each wavelength value in the reconstructed wavelength sequence of the previous sensing segment, the boundary correction is applied in a linearly decreasing manner according to the grating index position of each wavelength value from the boundary. The wavelength value closest to the boundary is given a full boundary correction, the wavelength value farthest from the boundary is given a zero correction, and the correction applied to the wavelength value in the middle position is determined by linear interpolation according to the position ratio of the wavelength value to the boundary. The boundary correction for the next sensing segment is applied in the same manner.

7. The method according to claim 6, characterized in that, In step two, after performing boundary continuity correction, the reconstructed wavelength sequence of each sensing segment after correction is used as the new initial value, and iterative hard threshold reconstruction is performed again for each sensing segment. When performing iterative hard threshold reconstruction again, the sparse representation basis is replaced with a local Fourier basis constructed with the reconstructed wavelength sequence after correction. The frequency range of each basis vector of the local Fourier basis is limited to the typical frequency range of wavelength changes caused by the deformation and leakage of the shield tunnel structure.

8. The method according to claim 1, characterized in that, In step two, the process of obtaining the strain distribution curve and temperature distribution curve based on the reconstructed wavelength distribution curve includes: setting several temperature reference gratings at intervals along the deployment path of the ultra-weak grating array sensing fiber; the temperature reference gratings are encapsulated with loose tubes to eliminate strain transmission so that the temperature reference gratings only respond to temperature changes; extracting the wavelength values ​​of each temperature reference grating position from the reconstructed wavelength distribution curve; calculating the temperature value of each temperature reference grating position based on the offset of the wavelength value of each temperature reference grating position relative to the nominal wavelength and the pre-calibrated temperature sensitivity coefficient; obtaining the temperature distribution curve by linear interpolation for the segments between adjacent temperature reference gratings; subtracting the wavelength offset component caused by temperature determined by the temperature distribution curve from the wavelength offset at each position in the reconstructed wavelength distribution curve to obtain the wavelength offset caused by pure strain; and converting the wavelength offset caused by pure strain into a strain distribution curve based on the pre-calibrated strain sensitivity coefficient.

9. The method according to claim 1, characterized in that, In step three, the joint determination of deformation and leakage includes: traversing the strain distribution curve and temperature distribution curve along the longitudinal direction of the shield tunnel using a sliding window of fixed length; calculating the difference between the maximum and minimum strain values ​​within each sliding window as the strain amplitude index for each sliding window; calculating the decrease in temperature value within each sliding window relative to the average temperature of adjacent sliding windows as the temperature negative bias index for each sliding window; when the strain amplitude index of a certain sliding window exceeds a preset strain anomaly threshold, the corresponding sliding window is marked as a deformation anomaly window; when the temperature negative bias index of a certain sliding window exceeds a preset temperature anomaly threshold, the corresponding sliding window is marked as a suspected leakage window; when a sliding window is simultaneously marked as both a deformation anomaly window and a suspected leakage window, the corresponding sliding window is marked as a leakage-deformation coupling window.

10. The method according to claim 9, characterized in that, In step four, the leakage-deformation coupling window generates a level one alarm, the suspected leakage window generates a level two alarm, and the abnormal deformation window generates a level three alarm. All sliding windows that generate alarms are sorted in descending order of alarm level, and the tunnel mileage locations corresponding to each sliding window that generates alarms are connected in sequence to generate an inspection and verification route and output to the monitoring terminal.

Images