Online health monitoring method and system based on tunnel construction structure

Abstract

The invention provides a tunnel construction structure health on-line monitoring method and system, and the method comprises the steps: obtaining stress data, crack images and seismic wave signals of tunnel construction, and carrying out the preprocessing; obtaining and constructing stress-crack joint representation mapping, generating a first joint feature vector, and predicting a lining health state of a future # imgabs0 # time step; generating a second joint feature vector, and determining a stratum stress field of the future # imgabs1 # time step according to the second joint feature vector; constructing a tunnel lining damage probability model according to the stress data and the crack image, and determining an overall construction risk score; and dividing the construction risks into a plurality of grades. According to the method, data multi-modal fusion, structure-stratum joint modeling and AI enhanced prediction and early warning are achieved, the defects in the prior art are effectively overcome, the method can be widely applied to engineering scenes such as tunnel construction, subway construction and underground space development, and the construction safety and the structure health monitoring capacity are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of online monitoring, and particularly relates to an online monitoring method and system for the structural health of tunnel construction. Background Art

[0002] During the tunnel construction process, the health status of the tunnel structure is directly related to the safety and stability of the project. Especially in complex geological conditions (such as soft surrounding rock, high in-situ stress area, fault zone, etc.), the tunnel lining, surrounding rock support, and ground stability face huge challenges. Currently, the engineering monitoring methods mainly include stress sensor monitoring, optical measurement, crack visual inspection, seismic wave detection, etc. These methods can provide the structural health status during tunnel construction to a certain extent, but there are still many deficiencies.

[0003] First, the limitations of the single-sensor monitoring method limit the comprehensiveness of monitoring. Traditional data acquisition means such as stress sensors and displacement sensors can only reflect the stress state of local structures and are difficult to comprehensively reflect the overall health of the entire tunnel. For example, during the tunnel excavation process, cracks may initiate and the surrounding rock may loosen in some areas. However, due to the limited density of sensor arrangement, it is often difficult to capture these tiny changes in a timely manner, resulting in the neglect of potential safety hazards.

[0004] Second, the disconnection between structural monitoring and ground instability monitoring makes it difficult for the early warning system to achieve accurate prediction. Existing tunnel monitoring systems usually process structural monitoring (such as lining deformation, crack propagation) and ground monitoring (such as microseismic detection, ground stress inversion) separately. In fact, the interaction between the tunnel support structure and the surrounding rock is a highly coupled non-linear dynamic process. For example, the stress change inside the ground may be manifested as crack propagation in the lining structure. However, traditional methods often ignore this dynamic connection and only rely on a single physical quantity for monitoring, resulting in a low prediction accuracy.

[0005] Third, there is a lack of intelligent prediction and early warning means. Existing methods mostly perform anomaly detection based on threshold determination or simple statistical methods and are difficult to identify potential structural risks in advance. For example, traditional deformation monitoring methods often calculate the structural stress through empirical formulas or simulation methods based on the finite element method (FEM). However, these methods have poor adaptability to complex environments (such as sudden changes in surrounding rock, local support failure) and cannot perform high-precision long-term trend prediction. In addition, the occurrence of ground instability is often sudden, and existing monitoring means lack the ability of active prediction. In some cases, even when the sensing data has shown stress anomalies, the system still cannot give an early warning with sufficient lead time before the accident occurs.

[0006] Finally, the insufficient ability of multimodal data fusion leads to information redundancy or loss among data. Since tunnel construction involves various types of sensor data (such as stress, cracks, seismic waves, etc.), traditional data processing methods often adopt an independent analysis approach and cannot effectively fuse data of different modalities. For example, when monitoring the crack development of a tunnel lining, relying solely on visual inspection may be interfered by factors such as lighting and occlusion. However, if stress sensing data is combined, the reliability of the crack propagation trend can be effectively improved. However, existing methods lack a unified cross-modal data fusion strategy, resulting in the inability to fully utilize information between different data sources and affecting the monitoring accuracy and prediction ability. Summary of the Invention

[0007] The object of the present invention is to design an online structural health monitoring method and system for tunnel construction. Through data multimodal fusion, structure-stratum joint modeling, AI-enhanced prediction and warning technologies, the deficiencies of the existing technology are effectively overcome, and it can be widely applied to engineering scenarios such as tunnel construction, subway construction, and underground space development, significantly improving construction safety and structural health monitoring capabilities.

[0008] To achieve the above object, in the first aspect of the present invention, an online structural health monitoring method for tunnel construction is provided. The method includes:

[0009] S1. Obtain stress data, crack images, and seismic wave signals of tunnel construction and perform preprocessing;

[0010] S2. Obtain the preprocessed stress data and crack images, construct a stress-crack joint characterization mapping, generate a first joint feature vector, and predict the lining health state at the future T time steps according to the first joint feature vector; wherein, predicting the lining health state at the future T time steps includes generating corresponding future stress data and future crack propagation images;

[0011] S3. Generate a second joint feature vector according to the preprocessed seismic wave signals and future stress data, and determine the formation stress field at the future T time steps according to the second joint feature vector to ensure that the prediction results conform to the wave physical law;

[0012] S4. Construct a structure-crack-formation three-modal information interaction network according to the predicted lining health state at the future T time steps and the formation stress field at the future T time steps to initialize the overall construction risk score. At the same time, construct a tunnel lining failure probability model according to the stress data and crack images to obtain the failure probability, and determine the overall construction risk score according to the tunnel lining failure probability model in combination with the future stress data;

[0013] S5. Construct an adaptive multi - level early warning mechanism for the overall construction risk score at the future T time step, set the early warning threshold, divide the construction risk into multiple levels, and dynamically optimize the early warning threshold in combination with reinforcement learning to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk.

[0014] Preferably, the pre - processing includes:

[0015] Collect stress data and perform dynamic correction of the historical mean of the sliding window to obtain the dynamically corrected stress data;

[0016] Use a local adaptive enhancement method based on adaptive histogram equalization to enhance the crack image to make the crack edges clearer;

[0017] Use a micro - seismic sensor array to collect seismic wave data in real - time, form a time - series signal, and denoise the time - series signal to obtain the denoised seismic wave signal;

[0018] Perform dynamic time warping based on the dynamically corrected stress data, the enhanced crack image, and the denoised seismic wave signal to synchronize the time of all data sources.

[0019] Preferably, the dynamic time warping finds the optimal time alignment point through the minimum time deviation loss; where the time deviation loss is expressed as:

[0020]

[0021] where represents the loss function of time alignment; g(W t ′) represents the mapping function from the seismic wave signal to stress; h(S t ′) represents the mapping function from stress data to crack propagation; S t ′ represents the corrected stress data, and I t ′ represents the enhanced crack image.

[0022] Preferably, the first joint feature vector is fused using a stress - crack cross - attention mechanism, so that the force change in the crack area can reversely affect the stress field modeling, improve the spatio - temporal coupling, and correct the data of each stress sensor point based on the deformation gradient in the crack area.

[0023] Preferably, predicting the lining health state at the future T time step according to the first joint feature vector includes:

[0024] A recursive prediction model based on a spatio-temporal autoregressive unit is used to calculate the stress and crack state in the future T time steps. Meanwhile, an influence term of the local stress field of crack propagation is introduced into the recursive prediction model to correct the recursive prediction model, and a regularization optimization term based on a crack propagation gradient regularization term and a stress prediction smoothing regularization term is designed to optimize the recursive prediction model, and the health state of the tunnel lining in the future T time steps is output.

[0025] Preferably, the second joint feature vector contains information of seismic wave signals and predicted stress states. The second joint feature vector is subjected to second joint feature extraction mapping by a spatio-temporal joint filtering layer to fuse the seismic wave signals and the stress prediction results. Meanwhile, on the basis of the second joint feature extraction mapping, combined with the influence of structural stress data, an improved wave-stress joint feature constraint term is defined to make the formation stress prediction conform to the wave physical law; the improved wave-stress joint feature constraint term is expressed as:

[0026]

[0027] Wherein, represents the loss function of the coupling of the wave equation and the stress distribution; represents the spatial gradient of the seismic wave signal, indicating the propagation mode of the wave information in the formation; α represents the stress-wave coupling coefficient, controlling the influence intensity of the seismic wave on the formation stress; represents the predicted future stress data, which is used to adjust the wave characteristics and improve the inversion accuracy;

[0028] Then, determining the formation stress field in the future T time steps according to the second joint feature vector includes:

[0029] Based on the future stress data and seismic wave signals, a physical constraint neural network is used for stress inversion, so that the predicted formation stress field not only conforms to the data-driven result, but also satisfies the wave equation constraint. Specifically:

[0030] Seismic wave propagation equation:

[0031]

[0032] Wherein, u represents the seismic wave displacement component; c represents the propagation speed of the seismic wave in the formation, which is related to the formation stress and material properties; f(x,t) represents the source function, reflecting the external excitation of the seismic wave;

[0033] Combined with the loss function of the physical constraint neural network training, the predicted formation stress field satisfies the seismic wave physical constraint:

[0034]

[0035] Among them, represents the physical consistency loss, ensuring that the stress field inverted by the AI satisfies the seismic wave propagation equation and improving the prediction credibility.

[0036] Preferably, the tunnel lining failure probability model is determined based on the failure probability function, and the failure probability function is calculated as follows:

[0037]

[0038] where P fail represents the lining structure failure probability, indicating the risk that the tunnel lining may fail within the next T time steps; σ max represents the ultimate bearing stress of the lining material; η represents the stress influence coefficient, measuring the influence of stress overload on the failure probability; λ represents the crack influence coefficient, measuring the influence of the crack propagation rate on the failure probability; Crack propagation gradient, representing the crack growth rate, enabling the failure probability to consider the crack propagation trend;

[0039] The overall construction risk score is expressed as:

[0040]

[0041] where R t represents the comprehensive construction risk score, indicating the risk level in the next T time steps; α represents the structure failure weight, controlling the contribution of P fail to the risk score; β represents the formation stress influence weight, measuring the influence of abnormal formation stress on the risk score.

[0042] Preferably, an adaptive multi-level early warning mechanism is constructed for the overall construction risk score for the next T time steps, and the early warning threshold is set. The construction risk is divided into multiple levels by using a non-linear piecewise function, including:

[0043] Low risk: Normal construction, no adjustment required;

[0044] Medium risk: Local construction adjustment, increasing the support strength;

[0045] High risk: Immediately stop work and carry out emergency support.

[0046] Preferably, the early warning threshold is dynamically optimized by combining reinforcement learning to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk, including:

[0047] The reinforcement learning is used to intelligently adjust the early warning threshold, and the reward function is defined to ensure that the early warning system can avoid false alarms while reducing the accident rate, and the time difference is used to update the early warning threshold:

[0048]

[0049] Among them, R actual represents the risk score at the time of an actual accident; represents the safety improvement index after construction adjustment, measuring the effectiveness of early warning; α and β represent adjustment coefficients to ensure an optimized balance between false alarms and missed alarms; R safe represents the safety value, and R actual represents the actual value;

[0050] Combining historical construction data with the current risk status, define an optimized support parameter set, and the construction adjustment strategy of the optimized support parameter set meets the engineering requirements. Then, use the optimal adjustment strategy selection to calculate the degree of risk reduction, and select the optimal plan from the optimized support parameter set.

[0051] In the second aspect of the present invention, an on-line health monitoring system for tunnel construction structures is provided, and the system includes:

[0052] A tunnel construction data acquisition unit for acquiring stress data, crack images, and seismic wave signals of tunnel construction and performing preprocessing;

[0053] A stress-crack analysis unit for acquiring the preprocessed stress data and crack images, constructing a stress-crack joint characterization mapping, generating a first joint feature vector, and predicting the lining health state at the next T time steps according to the first joint feature vector; wherein, the prediction of the lining health state at the next T time steps includes generating corresponding future stress data and future crack propagation images;

[0054] A formation stress field prediction unit for generating a second joint feature vector according to the preprocessed seismic wave signals and future stress data, and determining the formation stress field at the next T time steps according to the second joint feature vector to ensure that the prediction result conforms to the wave physics law;

[0055] A risk prediction unit for constructing a structure-crack-formation three-modal information interaction network to initialize the overall construction risk score according to the predicted lining health state at the next T time steps and the formation stress field at the next T time steps, and simultaneously constructing a tunnel lining failure probability model according to the stress data and crack images to obtain the failure probability, and determining the overall construction risk score according to the tunnel lining failure probability model combined with the future stress data;

[0056] A health early warning unit for constructing an adaptive multi-level early warning mechanism for the overall construction risk score at the next T time steps, setting an early warning threshold, dividing the construction risk into multiple levels, and dynamically optimizing the early warning threshold in combination with reinforcement learning to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk.

[0057] The beneficial technical effects of the present invention are at least as follows:

[0058] (1) Traditional methods are difficult to process stress data (numerical sequences) and crack images (visual data) simultaneously. The present invention innovatively adopts a dual-modal Transformer, combines structural stress data and crack propagation information, and models the dynamic correlation between the two through an adaptive attention mechanism (Cross-Attention), making up for the problem of limited monitoring range of single-sensor data.

[0059] This method can not only predict the future structural stress distribution but also infer the evolution trend of cracks, realizing the all-round monitoring of the health state of tunnel lining structures.

[0060] (2) Existing formation monitoring methods are mainly based on finite element calculations or empirical formulas and are difficult to track the dynamic changes of formation stress in real time. The present invention collects seismic wave data through microseismic sensors and combines it with a physics-informed neural network (PINN) for seismic wave inversion to accurately calculate the stress field distribution inside the formation.

[0061] This method breaks through the computational bottleneck of traditional formation monitoring, enabling the system to monitor the instability trend of the formation in real time and give early warnings of potential collapse risks, solving the problem that existing methods are difficult to predict sudden formation changes in advance.

[0062] (3) Traditional methods separately process structural monitoring (stress, cracks) and formation monitoring (seismic waves), while the present invention proposes a global risk prediction model that combines the structural health state (cracks and stresses predicted by the Transformer) and the formation stress state (stress field inverted from seismic waves) to establish a multi-modal spatio-temporal prediction network.

[0063] By fusing the characteristic information of different data sources, the present invention can more accurately predict the overall risk state at the future T time step, greatly improving the warning accuracy and avoiding false alarms and missed alarms. Description of the Drawings

[0064] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the following drawings without creative efforts.

[0065] Figure 1 It is a flowchart of the online structural health monitoring method for tunnel construction according to the present invention.

[0066] Figure 2 It is a framework diagram of the online structural health monitoring system for tunnel construction according to the present invention. Detailed Embodiments

[0067] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

[0068] In one or more embodiments, as Figure 1 shown, an online structural health monitoring method for tunnel construction is disclosed, and the method includes the following steps S1 - S5:

[0069] S1. Obtain stress data, crack images, and seismic wave signals of tunnel construction and perform preprocessing.

[0070] Specifically, the objective of this step is to obtain stress data, crack images, and seismic wave signals and perform high-precision preprocessing specifically for tunnel construction health monitoring to provide reliable inputs for subsequent structural health prediction and formation stress inversion. Due to the complex tunnel construction environment, the data quality is affected by factors such as environmental noise, sensor drift, non-synchronization, and crack detection difficulty, and a series of processes are required to ensure the stability and availability of the data.

[0071] Furthermore, collect S t where S t ={s1, s2,..., s n}, representing n groups of stress data at time t, and each s i comes from sensors at different positions.

[0072] Due to the large variation in the construction environment, the sensors may produce systematic drift (such as zero offset caused by temperature change), and a dynamic correction model is defined:

[0073]

[0074] where s i ′ is the corrected stress data. α is the historical correction coefficient, dynamically adjusted based on the historical mean of the sliding window to reduce long-term deviation. λ is the adaptive adjustment coefficient, enabling the correction to reduce oscillations while quickly responding to mutations. is the change rate of the stress data, used for regularization with an additional innovation term to avoid the influence of mutations and improve measurement stability.

[0075] Furthermore, arrange high-definition crack cameras on the lining surface and collect a set of crack images every Δt, denoted as I t ={I1, I2,..., I m}, where each I i is a single-frame image.

[0076] Due to the uneven illumination in the tunnel construction environment, a local adaptive enhancement method is adopted to make the crack edges clearer:

[0077]

[0078] where I i ′ is the enhanced crack image. β is the contrast enhancement coefficient, calculated based on Adaptive Histogram Equalization (AHE) to optimize the crack contrast under different brightness levels. γ is the Laplacian enhancement coefficient, controlling the additional innovation term. Calculates the second-order gradient of the image using the Laplacian operator to enhance the crack boundary details, making the crack edges more obvious and reducing false detections.

[0079] Furthermore, a microseismic sensor array is arranged around the tunnel to collect seismic wave data in real-time, forming a time series signal W t , where:

[0080] W t = {w1, w2,..., w k}}, and each w i represents the waveform signal at a certain moment.

[0081] Since the seismic wave signal contains effective waves (surrounding rock stress waves) + noise (equipment vibration, wind noise, etc.), noise reduction processing is required:

[0082]

[0083] where w i ′ is the noise-reduced seismic wave signal. φ j (w i ) is the j-th orthogonal basis component of the seismic wave, representing the main mode of the signal. γ j is the weight of the basis component. Only the top p-dimensional components with the highest energy are retained to make the seismic wave signal better match the stress changes inside the surrounding rock. η is the second-order gradient noise reduction coefficient, controlling the additional innovation term. Calculates the high-frequency components of the vibration for removing irrelevant noise.

[0084] Furthermore, due to the different timestamps of different data sources, an innovative multi-modal time alignment method is adopted to synchronize the time of all data sources:

[0085]

[0086] where, is the time alignment mapping function, based on Dynamic Time Warping (DTW), finding the optimal time alignment point through the minimum time deviation loss. An autoregressive weight term is additionally added to make the matching between data consider not only time but also optimize the physical correlation:

[0087]

[0088] Among them, is the loss function for time alignment. g(W t ′) is the mapping function from seismic wave signals to stress. h(S t ′) is the mapping function from stress data to crack propagation.

[0089] S2. Obtain the preprocessed stress data and crack images, construct a stress-crack joint characterization mapping, generate the first joint feature vector, and predict the lining health state at the next T time steps according to the first joint feature vector; among them, predicting the lining health state at the next T time steps includes generating the corresponding future stress data and future crack propagation images.

[0090] Specifically, the goal of this step is to build a model for predicting the stress state and crack evolution trend of the lining within the next T time steps based on the corrected stress data output in step S1 and the enhanced crack images. Since the stress field and crack distribution of the tunnel lining are highly coupled, simply independently modeling the two may lead to prediction distortion. This step proposes a stress-crack joint spatio-temporal modeling method, using a stress-crack dual-modal fusion network to add the stress feedback of the crack area on the basis of stress time series modeling to improve the prediction accuracy.

[0091] Furthermore, input the corrected stress data representing the stress states of n sensor points on the tunnel lining.

[0092] The enhanced crack images represent the lining crack images.

[0093] Among them, since the stress data is a time series signal and the crack images are spatial data, a mapping relationship must be constructed so that the model can extract the key stress effects from the morphological features of the cracks.

[0094] Define the stress-crack joint characterization mapping to establish the relationship between the two:

[0095]

[0096] Among them, h t is the first joint feature vector, which fuses the stress state and crack distribution information. is the joint encoding function, which adopts a stress-crack cross-attention mechanism to enable the stress changes in the crack area to reversely affect the stress field modeling and improve the spatio-temporal coupling.

[0097] Among them, the calculation method:

[0098] For each stress sensor point s i , find its corresponding crack area Calculate the deformation gradient of the crack area Use as the weight to correct s i , and form the crack influence corrected stress

[0099]

[0100] Among them, β is the crack deformation influence coefficient, which determines the influence weight of the crack gradient on the stress. is the gradient amplitude of the crack area, indicating the degree of influence of crack propagation on the local stress.

[0101] Furthermore, set the goal to predict the lining health state at the next T time steps, that is, the future stress and crack state

[0102] Then there is a modeling framework:

[0103] Adopt a recursive prediction model to calculate the stress and crack state at the next T time steps:

[0104]

[0105] Among them, is the predicted future stress state. is the predicted future crack state. is the recursive prediction model, which adopts a Spatio-Temporal Recurrent Unit (STRU), enabling the historical stress and crack distribution to jointly act on the future prediction.

[0106] Furthermore, since crack propagation often has sudden changes, introduce the local stress field influence term of crack propagation to correct the prediction model:

[0107]

[0108] Among them, is the prediction error loss function, ensuring that the future stress prediction conforms to the crack evolution trend. is the local stress distribution calculated based on crack propagation, serving as a constraint term for stress prediction.

[0109] Furthermore, to improve the stability of the prediction and avoid overfitting and unreasonable crack morphology prediction. Set the regularization optimization term:

[0110]

[0111] Among them, is the crack propagation gradient regularization term to prevent the distortion of the predicted crack shape. is the stress prediction smoothing regularization term to prevent non-physical violent fluctuations in the prediction results.

[0112] The final output of this step is to predict the health state of the tunnel lining in the next T time steps:

[0113] Predicted future stress data for step 4 to conduct risk assessment.

[0114] Predicted future crack propagation images used to evaluate whether the crack exceeds the safety threshold.

[0115] S3. According to the preprocessed seismic wave signals and future stress data, generate the second combined feature vector, and determine the formation stress field in the next T time steps based on the second combined feature vector to ensure that the prediction results conform to the wave physics law.

[0116] Specifically, the goal of this step is based on the denoised seismic wave data output by step S1 and the future stress data predicted in step S2 to construct a formation stress inversion model and calculate the formation stress field within the next T time steps During the tunnel construction process, the formation stress of the surrounding rock is affected by factors such as excavation disturbance, lining support, and geological heterogeneity, resulting in a complex and dynamically changing stress field distribution. Traditional seismic wave inversion methods (such as finite element inversion) have high computational costs and are difficult to accurately capture the evolution of complex formation stress. This step combines a data-driven and physically constrained AI inversion model to invert the formation stress through seismic wave data and uses the joint constraint of structural stress-seismic wave to improve the accuracy and stability of the prediction.

[0117] Furthermore, the joint modeling of seismic wave data and structural stress

[0118] Input:

[0119] Denoised seismic wave data representing k groups of seismic wave signals collected at time t (from step 1).

[0120] Predicted future stress state representing the lining stress in the next T time steps (from step 2).

[0121] Furthermore, due to the fact that the propagation characteristics of seismic waves are affected by formation stress and material distribution, when inverting the formation stress field, it is necessary to simultaneously consider the coupling relationship between seismic wave data and structural stress state.

[0122] Adopt a seismic wave propagation feature extraction function Calculate the wave characteristics related to the formation stress field:

[0123]

[0124] Among them, H t is the second combined feature vector, containing information on seismic wave signals and predicted stress states. is the combined feature extraction mapping. By using a spatio-temporal joint filtering layer, the seismic wave signal and the stress prediction result are fused, enabling the seismic wave information to more effectively reflect the formation stress state.

[0125] On this basis, an improved wave-stress combined feature constraint term is defined to make the formation stress prediction conform to the wave physical law and simultaneously consider the influence of structural stress:

[0126]

[0127] Among them, is the loss function for the coupling of the wave equation and the stress distribution. is the spatial gradient of the seismic wave signal, representing the propagation mode of wave information in the formation. α is the stress-wave coupling coefficient, controlling the influence intensity of seismic waves on the formation stress. is the future lining stress state predicted in step 2, used to adjust the wave characteristics and improve the inversion accuracy.

[0128] Furthermore, the inversion of the formation stress field based on the Physics-Informed Neural Network (PINN)

[0129] Based on seismic wave data and stress prediction Calculate the formation stress field at the future T time steps Ensure that the prediction result conforms to the wave physical law.

[0130] Use the Physics-Informed Neural Network (PINN) for stress inversion, making the predicted formation stress field not only conform to the data-driven result but also satisfy the wave equation constraint.

[0131] Seismic wave propagation equation (controlling the evolution of the formation stress field):

[0132]

[0133] Among them, u is the seismic wave displacement component. c is the propagation speed of seismic waves in the formation, related to the formation stress and material properties. f(x, t) is the source function, reflecting the external excitation of seismic waves.

[0134] Combined with the loss function for PINN training, making the predicted formation stress field Meet the physical constraints of seismic waves:

[0135]

[0136] Among them, is the physical consistency loss, ensuring that the stress field inverted by AI satisfies the seismic wave propagation equation and improving the prediction credibility.

[0137] Furthermore, in order to prevent the predicted formation stress field from having unreasonable mutations and improving the prediction stability.

[0138] Adopt a stress balance regularization term to optimize the formation stress so that it conforms to historical trends and physical laws:

[0139]

[0140] Among them, is the stress balance loss, ensuring that the formation stress evolves smoothly over time. is the predicted future formation stress field. is the formation stress field at the previous time step. γ is the stress balance regularization coefficient, controlling the temporal consistency of the prediction results and preventing sudden changes in the prediction results.

[0141] The final output of this step is to predict the formation stress state at future T time steps:

[0142] The predicted future formation stress field is used for risk assessment in Step 4.

[0143] The formation stress distribution that conforms to physical constraints ensures that the seismic wave inversion results are physically reasonable.

[0144] S4. According to the predicted lining health state at future T time steps and the formation stress field at future T time steps, construct a three-modal information interaction network of structure-crack-formation to initialize the overall construction risk score. At the same time, construct a tunnel lining failure probability model based on stress data and crack images to obtain the failure probability, and determine the overall construction risk score according to the tunnel lining failure probability model combined with future stress data.

[0145] Specifically, the goal of this step is to construct a comprehensive tunnel structure safety risk assessment method based on the predicted future structural health state in Step 2 and the predicted future formation stress field calculated in Step 3 This step proposes a structure-formation coupling risk modeling to calculate the overall safety state of the tunnel at future T time steps.

[0146] Furthermore, joint risk feature extraction:

[0147] Input: Structural health prediction data It represents the stress states of each measuring point of the lining structure at the future T time steps. Crack propagation prediction data It represents the crack images at the future T time steps. Formation stress field It represents the formation stress distribution at the future T time steps.

[0148] Construct a multi-modal risk feature vector:

[0149] Due to the high coupling relationship among structural stress, crack propagation, and formation stress state, a joint risk feature extraction mapping Calculate the initial overall construction risk state:

[0150]

[0151] Among them, r t is the initial overall construction risk score, comprehensively representing the tunnel safety state within the future T time steps. is the joint feature mapping function, adopting a structure-crack-formation three-modal information interaction network to ensure that the model can accurately evaluate the impact of each factor on tunnel health.

[0152] Furthermore, due to the dynamic changes in risk features at different time steps, a time-dependent term is introduced so that the risk features can reflect the risk evolution trend:

[0153] R t = R t + γ·|R t - R t-1 || (17)

[0154] Among them, γ is the dynamic risk strengthening coefficient, controlling the sensitivity of risk feature changes. ||R t - R t-1 || is to measure the difference in risk features between the current time step and the previous time step, ensuring the stability of risk assessment results.

[0155] Furthermore, the goal is to quantify the probability of failure of the lining structure within the future T time steps to support safety decisions. Define the failure probability function:

[0156] Adopt stress state and crack characteristics to construct a tunnel lining failure probability model:

[0157]

[0158] Among them, P fail is the lining structure failure probability, representing the risk that the tunnel lining may fail within the future T time steps. σ maxis the ultimate bearing stress of the lining material. η is the stress influence coefficient, which measures the impact of stress overloading on the failure probability. λ is the crack influence coefficient, which measures the impact of crack propagation rate on the failure probability. is the crack propagation gradient, representing the crack growth rate, enabling the failure probability to consider the crack propagation trend.

[0159] Furthermore, combining the formation stress state calculate the overall construction risk score R t , providing a quantitative index for the construction safety level. Define the comprehensive risk score:

[0160] Calculate the risk score using the stress state, crack evolution, and formation stability:

[0161]

[0162] where R t is the comprehensive construction risk score, representing the risk level in the future T time steps. α is the structural failure weight, controlling the contribution of P fail to the risk score. β is the formation stress influence weight, measuring the impact of abnormal formation stress on the risk score. E safe is the safety threshold of the formation stress, ensuring that the risk score can accurately quantify the risk of formation instability.

[0163] Furthermore, to prevent short-term abnormal fluctuations in the risk score and improve the stability of risk assessment. Introduce the risk smoothing regularization term:

[0164] Since the risk score R t may be affected by local measurement errors or short-term stress fluctuations, define the risk smoothing regular term:

[0165]

[0166] where is the risk smoothing loss, controlling the smooth change of the risk score over time. δ is the smoothing regularization weight, controlling the smoothing degree, ensuring that the risk score will not be misjudged due to short-term abnormal changes.

[0167] The final output of this step is the comprehensive risk assessment index, which is used in Step 5 for intelligent early warning:

[0168] The failure probability P fail , characterizing the failure risk of the future tunnel lining.

[0169] The construction risk score R t , providing a quantitative tunnel safety level.

[0170] S5. Construct an adaptive multi - level early warning mechanism for the overall construction risk score at the future T time step, set the early warning threshold, divide the construction risk into multiple levels, and dynamically optimize the early warning threshold in combination with reinforcement learning to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk.

[0171] Specifically, the goal of this step is based on the comprehensive construction risk score R calculated in step S4 t , construct an intelligent early warning system, and dynamically optimize the early warning threshold R safe , to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk. This step adopts an adaptive multi - level early warning mechanism, combines historical risk trends and real - time risk assessments, optimizes construction decisions, and improves construction safety.

[0172] Furthermore, set the multi - level early warning rules: Based on the construction risk score R t , representing the comprehensive safety risk at the future T time step. Historical risk data R t-1 , R t-2 ,... are used to dynamically adjust the early warning threshold.

[0173] Set the adaptive early warning strategy:

[0174] Use a non - linear piece - wise function to divide the construction risk into multiple levels, and define the early warning threshold R safe :

[0175]

[0176] Among them, is the mean of the historical risk scores, used as the basic value for setting the early warning threshold. σ R is the standard deviation of the risk scores, measuring the risk volatility to ensure that the threshold adapts to dynamic changes. θ1, θ2 are weight coefficients, dynamically adjusted according to construction conditions and risk trends.

[0177] Furthermore, risk level division:

[0178] Low risk (R t < R safe,low ): Normal construction, no adjustment required.

[0179] Medium risk (R safe，low ≤R t <R safe，high ): Local construction adjustment, increase the support strength.

[0180] High risk (R t ≥R safe，high ): Immediately stop work and carry out emergency support.

[0181] Furthermore, use reinforcement learning to dynamically optimize R safe, reduce false alarms and missed alarms, and improve the accuracy of early warnings. Construction of optimization strategies:

[0182] Adopt reinforcement learning to intelligently adjust the early warning threshold and define the reward function Ensure that the early warning system reduces the accident rate while avoiding false alarms:

[0183]

[0184] Among them, R actual is the risk score when a real accident occurs, is the safety improvement index after construction adjustment, measuring the effectiveness of early warnings. α and β are adjustment coefficients to ensure an optimal balance between the impacts of false alarms and missed alarms.

[0185] Update strategy:

[0186] Adopt the temporal difference update rule:

[0187] R safe ←R safe +δ(R actual -R safe ) (23)

[0188] Among them, δ is the learning rate, controlling the dynamic adjustment range of the early warning threshold.

[0189] Furthermore, in medium and high-risk situations, intelligently recommend construction adjustment plans to improve the tunnel support capacity.

[0190] Calculation of adjustment plan:

[0191] Combine historical construction data with the current risk status and define the set of optimized support parameters Ensure that the construction adjustment strategy meets the engineering requirements:

[0192]

[0193] Among them, A i is the specific construction adjustment strategy, including adding support, adjusting the excavation sequence, optimizing the drainage plan, etc. Adopt the optimal adjustment strategy selection, calculate the degree of risk reduction, and select the optimal plan:

[0194]

[0195] Among them, A * is the optimal construction adjustment plan. R t (A i ) is the risk score after implementing the adjustment plan A i .

[0196] The final output of this step is the intelligent early warning signal and construction adjustment suggestions:

[0197] Intelligent warning level (low / medium / high risk), used for construction safety management.

[0198] Optimized warning threshold R safe , ensuring the long-term stable operation of the warning system.

[0199] Optimal construction adjustment plan A * , providing the optimal support optimization strategy.

[0200] In one or more embodiments, as Figure 2 shown, an online monitoring system for the structural health of tunnel construction is disclosed, characterized in that the system includes:

[0201] Tunnel construction data acquisition unit 301, used to acquire stress data, crack images, and seismic wave signals of tunnel construction and perform preprocessing;

[0202] Stress-crack analysis unit 302, used to acquire the preprocessed stress data and crack images, construct a stress-crack joint characterization mapping, generate a first joint feature vector, and predict the lining health state at the future T time steps according to the first joint feature vector; wherein, the prediction of the lining health state at the future T time steps includes generating corresponding future stress data and future crack propagation images;

[0203] Formation stress field prediction unit 303, used to generate a second joint feature vector according to the preprocessed seismic wave signals and future stress data, and determine the formation stress field at the future T time steps according to the second joint feature vector, ensuring that the prediction results conform to the wave physical law;

[0204] Risk prediction unit 304, used to construct a structure-crack-formation three-modal information interaction network according to the predicted lining health state at the future T time steps and the formation stress field at the future T time steps to initialize the overall construction risk score, and at the same time construct a tunnel lining failure probability model according to the stress data and crack images to obtain the failure probability, and determine the overall construction risk score according to the tunnel lining failure probability model combined with the future stress data;

[0205] Health warning unit 305, used to construct an adaptive multi-level warning mechanism for the overall construction risk score at the future T time steps, set a warning threshold, divide the construction risk into multiple levels, and dynamically optimize the warning threshold in combination with reinforcement learning to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk.

[0206] It should be noted that the specific working process of the online monitoring system for the structural health of tunnel construction provided in the embodiments of the present invention is the same as the process of the online monitoring method for the structural health of tunnel construction described in the above embodiments, and will not be elaborated here.

[0207] Compared with the prior art, the online monitoring system for the structural health of tunnel construction provided by the embodiment of the present invention acquires stress data, crack images, and seismic wave signals of tunnel construction and performs preprocessing; acquires the preprocessed stress data and crack images, constructs a stress-crack joint characterization mapping, generates a first joint feature vector, and predicts the lining health state at the future T time steps according to the first joint feature vector; wherein, predicting the lining health state at the future T time steps includes generating corresponding future stress data and future crack propagation images; generating a second joint feature vector according to the preprocessed seismic wave signals and future stress data, and determining the formation stress field at the future T time steps according to the second joint feature vector to ensure that the prediction result conforms to the wave physics law; constructing a structure-crack-formation three-modal information interaction network according to the predicted lining health state at the future T time steps and the formation stress field at the future T time steps to initialize the overall construction risk score, and at the same time constructing a tunnel lining failure probability model according to the stress data and crack images to obtain the failure probability, and determining the overall construction risk score according to the tunnel lining failure probability model combined with the future stress data; constructing an adaptive multi-level early warning mechanism for the overall construction risk score at the future T time steps to set the early warning threshold, dividing the construction risk into multiple levels, and dynamically optimizing the early warning threshold in combination with reinforcement learning to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk.

[0208] The embodiment of the present invention also provides an online monitoring device for the structural health of tunnel construction, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the embodiment of the online monitoring method for the structural health of tunnel construction as described above are implemented, such as Figure 1 the steps S1 to S5 described therein; or, when the processor executes the computer program, the functions of each module in the above system embodiments are implemented.

[0209] Exemplarily, the computer program can be divided into one or more modules. The one or more modules are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the online monitoring device for the structural health of tunnel construction.

[0210] The online monitoring device for the structural health of tunnel construction can be a computing device such as a desktop computer, notebook, palm computer, and cloud server. The online monitoring device for the structural health of tunnel construction may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the online monitoring device for the structural health of tunnel construction may further include input / output devices, network access devices, a bus, etc.

[0211] The processor can be a Central Processing Unit (CPU), or can also be other general-purpose processors, Digital Signal Processors (DSPs), Application-Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the online monitoring device for the structural health of tunnel construction, and connects various parts of the entire online monitoring device for the structural health of tunnel construction through various interfaces and lines.

[0212] The memory can be used to store the computer programs and / or modules. The processor realizes various functions of the online monitoring device for the structural health of tunnel construction by running or executing the computer programs and / or modules stored in the memory, and by calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc.; the data storage area can store data created according to the operation of the air-conditioning controller, etc. In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, at least one magnetic disk storage device, flash device, or other volatile solid-state storage devices.

[0213] Among them, if the module integrated with the online monitoring device for the structural health of tunnel construction is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0214] Those of ordinary skill in the art can understand that to implement all or part of the processes in the above-described embodiment methods, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-described method embodiments. Among them, the storage medium can be a magnetic disk, optical disc, read-only memory (ROM, Read-Only Memory), or random access memory (RAM, Random Access Memory), etc.

[0215] The above are the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements and refinements can still be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. An online health monitoring method for tunnel construction structures, characterized in that The method includes: S1. Obtain the stress data, crack images, and seismic wave signals of tunnel construction and perform preprocessing; S2. Obtain the preprocessed stress data and crack images, construct a stress-crack joint characterization mapping, generate a first joint feature vector, and predict the lining health state at the future T time steps according to the first joint feature vector; wherein, predicting the lining health state at the future T time steps includes generating corresponding future stress data and future crack propagation images; S3. Generate a second joint feature vector according to the preprocessed seismic wave signals and future stress data, and determine the formation stress field at the future T time steps according to the second joint feature vector to ensure that the prediction results conform to the wave physics law; S4. Construct a structure-crack-formation three-modal information interaction network according to the predicted lining health state at the future T time steps and the formation stress field at the future T time steps to initialize the overall construction risk score. At the same time, construct a tunnel lining failure probability model according to the stress data and crack images to obtain the failure probability, and determine the overall construction risk score according to the tunnel lining failure probability model combined with the future stress data; S5. Construct an adaptive multi-level early warning mechanism for the overall construction risk score at the future T time steps, set the early warning threshold, divide the construction risk into multiple levels, and dynamically optimize the early warning threshold in combination with reinforcement learning to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk.

2. The online health monitoring method for tunnel construction structures according to claim 1, wherein The preprocessing includes: Collect stress data and perform dynamic correction of the historical mean of the sliding window to obtain the dynamically corrected stress data; Use a local adaptive enhancement method based on adaptive histogram equalization to enhance the crack images to make the crack edges clearer; Use a microseismic sensor array to collect seismic wave data in real time to form a time series signal, and perform noise reduction on the time series signal to obtain the noise-reduced seismic wave signal; Perform dynamic time warping based on the dynamically corrected stress data, enhanced crack images, and noise-reduced seismic wave signals to synchronize the time of all data sources.

3. The online monitoring method for the structural health based on tunnel construction according to claim 2, characterized in that, The dynamic time warping finds the optimal time alignment point through the minimum time deviation loss; wherein, the time deviation loss is expressed as: Among them, represents the loss function for time alignment; g(W t ′) represents the mapping function from seismic wave signals to stress; h(S t ′) represents the mapping function from stress data to crack propagation; S t ′ represents the corrected stress data, and I t ′ represents the enhanced crack image.

4. The online monitoring method for the structural health of tunnel construction according to claim 1, characterized in that, The first joint feature vector is fused by using a stress-crack cross-attention mechanism, so that the stress change in the crack area can reversely affect the stress field modeling, improve the spatio-temporal coupling, and correct the data of each stress sensor point based on the deformation gradient of the crack area.

5. The online health monitoring method for tunnel construction structures according to claim 4, wherein The predicting the lining health state at the future T time steps according to the first joint feature vector includes: Use a recursive prediction model based on a spatio-temporal autoregressive unit to calculate the stress and crack states at the future T time steps. At the same time, introduce the local stress field influence term of crack propagation into the recursive prediction model to correct the recursive prediction model, and design a regularization optimization term based on the crack propagation gradient regularization term and the stress prediction smoothing regularization term to optimize the recursive prediction model, and output the lining health state of the tunnel at the future T time steps.

6. The online monitoring method for the structural health based on tunnel construction according to claim 5, wherein The second combined feature vector contains information on seismic wave signals and predicted stress states. The second combined feature extraction mapping is performed on the second combined feature vector using a spatio-temporal joint filtering layer to fuse the seismic wave signals and stress prediction results. Meanwhile, based on the second combined feature extraction mapping and considering the influence of structural stress data, an improved wave-stress combined feature constraint term is defined to ensure that the formation stress prediction conforms to wave physics laws. The improved wave-stress combined feature constraint term is expressed as: Among them, represents the loss function for coupling the wave equation and the stress distribution; represents the spatial gradient of the seismic wave signal, indicating the propagation mode of wave information in the formation; α represents the stress-wave coupling coefficient, controlling the influence intensity of seismic waves on formation stress; represents the predicted future stress data, used to adjust wave characteristics and improve inversion accuracy; Then, determining the formation stress field for the next T time steps based on the second combined feature vector includes: Based on the future stress data and seismic wave signals, a physics-constrained neural network is used for stress inversion, so that the predicted formation stress field not only conforms to the data-driven results but also satisfies the wave equation constraint. Specifically: Seismic wave propagation equation: where u represents the displacement component of the seismic wave; c represents the propagation speed of the seismic wave in the formation, which is related to the formation stress and material properties; f(x, t) represents the source function, which reflects the external excitation of the seismic wave; Combined with the loss function for training the physics-constrained neural network, the predicted formation stress field satisfies the physical constraints of seismic waves: Among them, represents the physical consistency loss, ensuring that the stress field inverted by AI satisfies the seismic wave propagation equation and improving the prediction credibility.

7. The online monitoring method for the structural health based on tunnel construction according to claim 6, wherein The tunnel lining failure probability model is determined based on the failure probability function, and the failure probability function is calculated as follows: Among them, P fail represents the failure probability of the lining structure, indicating the risk that the tunnel lining may fail within the next T time steps; σ max represents the ultimate bearing stress of the lining material; η represents the stress influence coefficient, measuring the impact of stress overload on the failure probability; λ represents the crack influence coefficient, measuring the impact of crack propagation rate on the failure probability; The crack propagation gradient represents the crack growth rate, enabling the failure probability to consider the crack propagation trend; The overall construction risk score is expressed as: Among them, R t represents the comprehensive construction risk score, indicating the risk level at the future T time steps; α represents the structural damage weight, controlling the contribution of P fail to the risk score; β represents the formation stress influence weight, measuring the impact of abnormal formation stress on the risk score.

8. The online health monitoring method for tunnel construction structures according to claim 1, characterized in that An adaptive multi-level early warning mechanism is set for the overall construction risk score for the next T time steps to set the early warning threshold. A non-linear piecewise function is used to set the early warning threshold to divide the construction risk into multiple levels, including: Low risk: Normal construction, no adjustment required; Medium risk: Local construction adjustment, increase the support strength; High risk: Immediately stop work and carry out emergency support.

9. The online monitoring method for the structural health based on tunnel construction according to claim 8, characterized in that Combining reinforcement learning to dynamically optimize the early warning threshold to ensure that the construction adjustment plan can accurately reduce the tunnel safety risk includes: Use reinforcement learning to intelligently adjust the warning threshold and define the reward function Ensure that the warning system reduces the accident rate while avoiding false alarms, and use temporal difference to update the warning threshold: Among them, R actual represents the risk score at the time of the actual accident; II(A) represents the safety improvement index after construction adjustment, measuring the effectiveness of early warning; α and β represent adjustment coefficients to ensure an optimized balance between false alarms and missed alarms; R safe represents the safety value, and R actual represents the actual value; Combining historical construction data and the current risk status, defining an optimized support parameter set. The construction adjustment strategy of the optimized support parameter set meets the engineering requirements. Then, the optimal adjustment strategy is selected to calculate the degree of risk reduction, and the optimal plan is selected from the optimized support parameter set.

10. An on-line health monitoring system for tunnel construction structures, characterized in that, The system includes: A tunnel construction data acquisition unit for acquiring stress data, crack images, and seismic wave signals of tunnel construction and performing preprocessing; A stress-crack analysis unit for acquiring the preprocessed stress data and crack images, constructing a stress-crack combined characterization mapping, generating a first combined feature vector, and predicting the lining health status for the next T time steps based on the first combined feature vector. Among them, predicting the lining health status for the next T time steps includes generating corresponding future stress data and future crack propagation images; A formation stress field prediction unit for generating a second combined feature vector based on the preprocessed seismic wave signals and future stress data, and determining the formation stress field for the next T time steps based on the second combined feature vector to ensure that the prediction results conform to wave physics laws; A risk prediction unit is used to construct a three-modal information interaction network of structure-crack-stratum to initialize the overall construction risk score based on the predicted lining health state in the next T time steps and the stratum stress field in the next T time steps. Meanwhile, a tunnel lining failure probability model is constructed based on stress data and crack images to obtain the failure probability, and the overall construction risk score is determined based on the tunnel lining failure probability model combined with future stress data. A health warning unit is used to construct an adaptive multi-level warning mechanism for the overall construction risk score in the next T time steps, set warning thresholds, divide the construction risks into multiple levels, and dynamically optimize the warning thresholds in combination with reinforcement learning to ensure that the construction adjustment plan can accurately reduce the tunnel safety risks.

Images