Tunnel rock-soil deformation identification method and system

Abstract

The application discloses a tunnel rock-soil deformation identification method and system, relates to the technical field of tunnel and underground engineering monitoring and early warning, and significantly improves the reliability and timeliness of tunnel soft surrounding rock deformation risk identification; in the sudden change response link, a time domain convolution network and a displacement acceleration adaptive reinforcement mechanism are introduced, so that the capturing ability of the system to nonlinear displacement surge is greatly improved; the application focuses on the acceleration component and applies physical constraint weight, ensures that the sudden change event is locked in real time, and avoids the missed report of accidents such as the instability of the working face; in view of the point cloud registration distortion problem in the joint development area, the geological law of rock mass structure surface is integrated into the registration algorithm, the normal elastic deformation threshold and the tangential coulomb friction criterion are coupled through the Gaussian mixture model, and the overall deformation misjudgment caused by the local displacement of rock blocks is inhibited from the root; the measurement accuracy of laser scanning in the fracture zone and the fault zone breaks through the environmental restriction, and provides a reliable basis for support decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel and underground engineering monitoring and early warning, in particular to a tunnel rock-soil deformation identification method and system. BACKGROUND

[0002] In tunnel engineering, the deformation monitoring of soft surrounding rock (such as mudstone, fracture zone) is directly related to construction safety; such rock mass is prone to nonlinear mutation after excavation disturbance, such as face instability or arch collapse; current engineering generally relies on real-time monitoring systems (such as three-dimensional laser scanning, optical fiber sensing) and intelligent early warning models to realize dynamic identification of deformation risk.

[0003] The identification scheme in the prior art is mainly based on two types of technology, one is to use long short-term memory network LSTM to analyze displacement sensor data and predict the deformation trend of surrounding rock; the other is to calculate the rock deformation by a multi-period laser scanning point cloud registration algorithm such as ICP; however, in the case of soft surrounding rock, the LSTM model relies on the continuity of historical data and responds slowly to sudden deformation of more than 200% of single-day displacement; and the point cloud registration is easily disturbed by local rock block movement in joint development rock mass, resulting in misjudgment of the overall deformation; although the optical fiber monitoring system can capture micro-strain, it cannot associate with the rock-soil physical yield mechanism to predict critical risks.

[0004] Recently, some schemes optimize data input or fuse multiple sources of sensors, such as increasing the laser scanning frequency to improve the point cloud density, or combining the vibrating string sensor to assist in judging the rock mass loosening; although such schemes can alleviate the noise interference of a single data source, they do not solve the core problem, and the extraction of the timing characteristics of the mutation event and the non-rigid deformation analysis of the joint rock mass still rely on the general algorithm framework, which is difficult to adapt to the complex mechanical behavior of soft surrounding rock; the incremental learning, transfer learning and other attempts to address the mutation miss report do not embed the physical boundary conditions of rock-soil damage, and the false positive rate is still high. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] The present application provides a tunnel rock-soil deformation identification method and system to solve the problem of high miss report rate of soft surrounding rock deformation caused by the lag of mutation response, distortion of joint registration and lack of micro-strain analysis in the prior art.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application provides a tunnel rock-soil deformation identification method, which comprises,

[0009] Step S1, obtain multi-period three-dimensional point clouds of the tunnel section by a multi-spectral laser scanner, obtain settlement / convergence time series data by a laser displacement meter, and obtain micro-strain time series data by a distributed optical fiber; align the data in time according to a unified clock, and complete coordinate unification and abnormal point elimination;

[0010] Step S2, input the displacement time series into a time domain convolution network to obtain displacement acceleration features including second-order time difference;

[0011] Step S3, calculate adaptive reinforcement weights based on displacement acceleration amplitude and noise scale estimation within a window to obtain a mutation feature reinforcement sequence;

[0012] Step S4, perform normal estimation and plane segmentation on the three-dimensional point cloud, and determine the main direction of the joint surface in combination with the field structure surface mapping / measurement line statistics; input the rock mass quality indicators including RQD / Jv as parameters of constraint strength or prior confidence;

[0013] Step S5, use non-rigid point cloud registration based on a Gaussian mixture model to register and solve the multi-period point cloud, and introduce a joint constraint term in the energy function to make the deformation field mainly slide in the tangential direction along the main direction of the joint, while satisfying:

[0014] The normal deformation does not exceed the threshold value determined by the elastic parameters;

[0015] The tangential sliding amount is limited by the Coulomb friction criterion;

[0016] Step S6, input the mutation feature reinforcement sequence, the point cloud deformation obtained based on registration, and the optical fiber micro-strain time series into a fusion decision module;

[0017] Step S7, analyze the state transition of the micro-strain feature by a hidden Markov state machine, and output a risk score or a critical risk signal;

[0018] Step S8, compare the risk score or the critical risk signal with the yield critical threshold value calculated according to the Mohr-Coulomb parameters, trigger a graded early warning, and output the result to a construction management system.

[0019] As a preferred scheme of the tunnel rock-soil deformation recognition method, the multi-spectral laser scanner simultaneously emits near-infrared and visible light bands and records the echo intensity, and performs dust point discrimination and point cloud completion based on the band intensity ratio and spatial neighborhood consistency.

[0020] As a preferred scheme of the tunnel rock-soil deformation recognition method, the adaptive reinforcement weights are calculated according to the absolute amplitude of the window second-order time difference and the robust noise scale, and are subjected to time continuity constraint to suppress isolated abnormal points, and the robust noise scale includes the median absolute deviation.

[0021] As a preferred scheme of the tunnel rock-soil deformation identification method, the joint main direction is determined by the statistical combination of the main plane set obtained by point cloud normal clustering and the strike / dip of the in-situ structural plane, and the RQD is used to adjust the normal constraint threshold or the constraint weight.

[0022] As a preferred scheme of the tunnel rock-soil deformation identification method, the non-rigid point cloud registration adopts a cooperative point drift or a non-rigid method based on a Gaussian mixture model, and a normal elastic threshold term and a tangential Coulomb friction term are added to the objective function to constrain the normal deformation and tangential slip of the joint surface.

[0023] As a preferred scheme of the tunnel rock-soil deformation identification method, the hidden Markov state machine includes three types of hidden states of elasticity-damage evolution-instability, the observation quantity is the amplitude, growth rate and frequency domain energy of the micro-strain, and the state transition probability can be updated online.

[0024] As a preferred scheme of the tunnel rock-soil deformation identification method, the hierarchical early warning includes at least three levels, and each level corresponds to different response suggestions and construction control parameter ranges.

[0025] In a second aspect, the present application provides a tunnel rock-soil deformation identification system, comprising,

[0026] A multi-source acquisition unit is configured to acquire multi-spectral three-dimensional point clouds, displacement time series and optical fiber micro-strain time series, and complete time synchronization.

[0027] A time series analysis module is configured to run a time domain convolution network and a sudden change feature enhancement calculation.

[0028] A point cloud processing module is configured to perform non-rigid registration based on a Gaussian mixture model with joint constraint and output deformation variables.

[0029] A fusion decision module is configured to output a risk score or a critical risk signal by using a hidden Markov state machine and a physical critical interface.

[0030] An early warning and visualization unit is configured to dynamically label and link the hierarchical early warning results and the joint sliding zone in a BIM / construction management model.

[0031] As a preferred scheme of the tunnel rock-soil deformation identification system, the physical critical interface receives a yield critical threshold calculated by the Mohr-Coulomb parameters obtained from indoor tests or numerical inversion, and supports online updating.

[0032] As a preferred scheme of the tunnel rock-soil deformation identification system, the pre-warning and visualization unit is configured to generate a grouting reinforcement range and process parameter suggestion according to a preset rule base and supply manual confirmation, and highlight display of a displacement mutation area and a joint sliding zone in the BIM model when a high risk level is reached.

[0033] The present application has the advantages that: the present application significantly improves the reliability and timeliness of tunnel soft surrounding rock deformation risk identification. In the mutation response link, the time domain convolution network and the displacement acceleration adaptive reinforcement mechanism are introduced, which greatly improves the system's ability to capture nonlinear displacement surge. The present application focuses on the acceleration component and applies physical constraint weight to ensure that the mutation event is locked in real time and avoid missing reports of accidents such as tunnel face instability. For the problem of point cloud registration distortion in joint development area, the geological law of rock mass structure surface is integrated into the registration algorithm, and the normal elastic deformation threshold and the tangential coulomb friction criterion are coupled by the Gaussian mixture model, which can effectively inhibit the overall deformation misjudgment caused by local rock block movement. The measurement accuracy of laser scanning in the fracture zone and fault zone breaks through the environmental restrictions and provides reliable basis for support decision.

[0034] In the implicit risk identification dimension, the scheme breaks through the static threshold limit of traditional optical fiber monitoring and builds a rock-soil damage evolution chain based on hidden Markov state mechanism. By mapping the micro-strain characteristics to the elastic-damage-instability three-state transition model, the system can capture the cumulative effect of continuous creep from the seemingly stable data stream and realize risk prediction hours in advance. This method does not require additional hardware, and the value of micro-strain data is deeply released only by algorithm reconstruction.

[0035] In addition, the whole process adopts a multi-source data fusion architecture: the outputs of the time series analysis module and the point cloud processing module are arbitrated by physical logic, and then drive the hierarchical early warning system to dynamically mark the high-risk area in the BIM model and automatically generate process parameter suggestions such as grouting range and anchor rod density. Not only does it reduce the dependence on expert experience, but it also compresses the construction response speed to the minute level, providing all-weather safety protection for complex geological tunnel construction. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope of the present application.

[0037] Figure 1 The figure is a flowchart of the tunnel rock-soil deformation identification method of the present application.

[0038] Figure 2 The figure is a framework diagram of the tunnel rock-soil deformation identification system of the present application. DETAILED DESCRIPTION

[0039] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0040] All the terms used in the present application (including technical and scientific terms) have the meanings generally understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted to have meanings consistent with the context of the present specification, and should not be interpreted in an idealized or overly formal manner.

[0041] For example, the terms "first", "second", etc. used in the present application are only used to distinguish the description of similar objects, to distinguish the first object from another object, and are not used to describe a specific order or sequence, nor can be understood to indicate or imply relative importance.

[0042] The present application proposes a tunnel geotechnical deformation identification method, which combines Figure 1 and Figure 2 As shown, the method comprises:

[0043] Step S1, acquiring multi-period three-dimensional point clouds of the tunnel section by a multi-spectral laser scanner, acquiring settlement / convergence time series data by a laser displacement meter, and acquiring micro-strain time series data by a distributed optical fiber; aligning the data in time according to a unified clock, and completing coordinate unification and abnormal point elimination;

[0044] Specifically, the unified clock refers to that the acquisition device records the time stamp with the same time source and checks the relative drift, the coordinate unification refers to that the external parameters of the laser scanner, the displacement meter and the optical fiber sensing are established corresponding to the target construction reference coordinate system, and the abnormal point elimination refers to that the dust noise points, reflection saturation points and isolated abnormal periods are robustly eliminated. Further, the conversion of point clouds to engineering coordinates is solved by the pose of the site control points or fixed targets, and the installation mileage and direction of the displacement meter and the optical fiber along the line are recorded according to the construction measurement and form a mapping table. In terms of numerical aperture, the scanning period is 15 min by default and can be adjusted to 10-30 min; the displacement meter sampling period is 10 s by default and can be adjusted to 1-60 s; the optical fiber sampling frequency is 20 Hz by default and can be adjusted to 5-50 Hz; the time alignment allows residual error to be ≤200 ms by default and can be adjusted to ≤500 ms, which is based on the cross-correlation peak width of the test run data and the device timing accuracy. Optionally, the timing method can use PPS hardware pulse or network time synchronization; coordinate unification can be performed once at the first deployment and after the device moves. Necessary abnormality and boundary processing includes: when a certain data stream is continuously missing for more than 5 sampling periods, the minimum executable aperture is to linearly extrapolate the nearest available segment and mark it as "degraded fusion", while keeping normal processing of the remaining data streams.

[0045] Step S2, inputting the displacement time series into the time-domain convolution network to obtain displacement acceleration features including second-order time difference;

[0046] Specifically, the time-domain convolution network is a one-dimensional convolution structure, the input is a time-ordered displacement scalar sequence, and the output is a multi-scale time series feature including first-order difference and second-order difference; the second-order time difference is calculated in the preprocessing stage or as a channel input in the network. Similarly, the receptive field is determined by the combination of the cavity convolution and the hierarchical stacking, which is used to cover the time span from short-term disturbance to daily trend. In terms of numerical aperture, the window length of feature extraction is 10 min by default and can be adjusted to 2-30 min; the convolution level is 3-5 layers by default with a step of 1; the inference sliding step is 1 sampling point by default; these settings are based on the trade-off between the recall of sudden transitions and the false alarm of steady-state segments on the validation set. Alternatively, in the limited computing resource scenario, an equivalent one-dimensional residual convolution network can be used instead, or a lightweight implementation containing only fixed difference kernel and sliding statistics can be used, while maintaining the consistency of the output feature dimension and order. When the input sequence length is shorter than the minimum window length, it is degraded to directly output the finite difference features and marked as "degraded features"

[0047] Step S3, based on the displacement acceleration amplitude and noise scale estimation within the window, the adaptive reinforcement weight is calculated to obtain the mutation feature reinforcement sequence;

[0048] Step S4, normal estimation and plane segmentation are performed on the three-dimensional point cloud, and the main direction of the joint surface is determined in combination with the site structural surface mapping / measurement line statistics; the rock mass quality indicators including RQD / Jv are used as parameter inputs to constrain the strength or prior confidence;

[0049] Similarly, normal estimation refers to obtaining the normal direction through local neighborhood point set least squares plane fitting, plane segmentation refers to extracting approximately coplanar point sets through region growing or random consistency strategy, and site structural surface mapping / measurement line statistics refers to using the strike, dip, and spacing information recorded by the working face or excavation face as directional prior. Alternatively, the neighborhood selection radius is 0.2-0.5 m by default or adaptive according to 20-50 points per neighborhood, the plane fitting residual threshold is 1-2 times the point spacing by default, and the direction consistency threshold is 10-20° by default. The setting basis is the point cloud density and the site joint roughness grade. Regarding parameter interpretation, when RQD and Jv are used as weight or threshold adjustment factors, the values are derived from core records and site measurement line statistics, and their role is to moderately relax the normal constraint or increase the prior uncertainty in poor quality areas. Necessary exception and boundary processing: when the local point density is insufficient or the dust obscures the normal, only the joint main direction in the high confidence area is output and the low confidence area is maintained in time sequence continuity using the previous period result.

[0050] Step S5, using non-rigid point cloud registration based on Gaussian mixture model to solve the registration of multi-period point clouds, and introducing a joint constraint term in the energy function, so that the deformation field is mainly tangential slip along the main direction of the joint, while satisfying:

[0051] The normal deformation does not exceed the threshold value determined by the elastic parameter;

[0052] The tangential slip is limited by the Coulomb friction criterion;

[0053] Step S6, input the mutation feature enhancement sequence, the point cloud deformation obtained by registration, and the optical fiber micro-strain time sequence into the fusion decision module; specifically, the input vector of the fusion decision module is aligned by timestamp, then dimensionless and abnormal duty cycle check, and in the case of single source missing, the weight decay method is used for continuous fusion to maintain the output stability. Optionally, the input update period is 5 minutes by default and can be adjusted to 1-10 minutes; the time alignment tolerance is 1 sampling step by default; the setting basis is the construction response time and the calculation resource limit. Regarding parameter interpretation, the setting of fusion threshold or weight is derived from the distribution fitting of historical samples or expert rules and can be dynamically adjusted according to the construction stage. When multiple sources are missing for more than one update period, the output enters the "hold" state and records the alarm "data cannot be judged" to avoid false triggering.

[0054] Step S7, analyze the state transition of micro-strain features through a hidden Markov state machine, and output risk scores or critical risk signals; further, the observation quantity of the hidden Markov state machine includes the normalized combination of micro-strain amplitude, slope and frequency band energy, the initial state probability and transition matrix are obtained by sample statistics of the stable and unstable segments in history, and are adjusted online in a sliding window during operation. Similarly, the minimum residence time is used to suppress frequent state switching, which is 2-5 observation steps by default; smoothing is realized by two equivalent methods of forward-backward probability average or Viterbi path backtracking. When the observation missing ratio exceeds 30%, the transition update is suspended and only the posterior most probable state of the previous time is output, and after recovery, a one-time re-estimation is performed.

[0055] Step S8, compare the risk score or critical risk signal with the yield critical threshold value calculated according to the Mohr-Coulomb parameters, trigger graded warning and output the result to the construction management system;

[0056] Specifically, the threshold of the hierarchical early warning is derived from the combination of indoor triaxial test parameters and field inversion results and reviewed by experts, and the risk level and response suggestion are issued in the construction management system through a data interface. Optionally, to avoid jitter, a time hysteresis mechanism and event merging rule are adopted: the same risk level needs to meet the minimum duration, and the interval between two triggers is less than the set air window and is merged into one event. Necessary boundary processing is: when the Mohr-Coulomb parameter interface is temporarily unavailable, the latest valid threshold is temporarily stored for comparison, and after recovery, it is automatically rechecked whether it needs to be upgraded or downgraded.

[0057] In one embodiment, the multi-spectral laser scanner simultaneously emits near-infrared and visible light bands and records the echo intensity, and discriminates dust points and completes point clouds based on the band intensity ratio and spatial neighborhood consistency;

[0058] In one embodiment, the adaptive emphasis weight is calculated according to the absolute amplitude of the window second-order time difference and the robust noise scale, and is subjected to time continuity constraint to suppress isolated abnormal points, and the robust noise scale includes the median absolute deviation;

[0059] Specifically, in step S3, the adaptive emphasis weight calculation process is as follows:

[0060] In each sampling time , the second-order time difference from step S2 is taken as the mutation sensitive quantity, and the robust noise scale is taken as the normalization reference, the normalized significance is first calculated and mapped to the weight, then the time continuity and morphological constraint are applied to the weight sequence, and finally the weight is multiplied by the second-order time difference to obtain the mutation feature enhancement sequence, which is used for subsequent fusion decision module, and the calculation formula includes:

[0061] The dimensionless adaptive emphasis weight is obtained by using a monotonic S-shaped mapping:

[0062] ,

[0063] Wherein, represents the adaptive emphasis weight at time , dimensionless, taking value , is the mapping slope coefficient, dimensionless, used to adjust the transition steepness, is the second-order time difference amplitude at time , consistent with the displacement, is the robust noise scale in the window, consistent with the displacement, is a small positive number to prevent the denominator from tending to zero, consistent with the displacement, taking value times the acquisition range, is the significance center threshold, dimensionless, used to control the alarm level;

[0064] where the second-order time difference is approximated by central difference:

[0065] ,

[0066] where, as before, , , are the values of the displacement time series at indices , , , with the same dimension as displacement,

[0067] The robust noise scale is estimated by MAD:

[0068] , ,

[0069] where, is the Gaussian approximation calibration constant , is the median operation on the index set , is the sliding window index set centered at , denotes the index variable within the window , with the same domain as i, is the first-order difference of displacement, with the same dimension as displacement, are the values of the displacement time series at indices , with the same dimension as displacement;

[0070] After obtaining the original sequence of , a first-order recursive time continuity constraint is applied to suppress isolated spikes, followed by a gray-scale morphological close-open process to fill in short gaps and filter out short responses, resulting in a smooth and structurally consistent weight sequence. Subsequently, this sequence is multiplied point-by-point with the second-order time difference to form a mutation feature enhancement sequence, which enters the subsequent fusion decision and early warning link;

[0071] Specifically, the second-order time difference is used as the mutation sensitivity quantity, combined with the robust noise scale to construct a dimensionless significance, and then a monotonic S-shaped mapping is used to obtain a weight between zero and one, thereby maintaining comparability under different noise levels and dimensional conditions. Time continuity processing makes the response naturally connected on the time axis, avoiding excessive response to occasional spikes, and morphological close-open further suppresses fragmented short segments and connects real events cut off by noise, making the obtained weight interval closer to the actual duration of surrounding rock deformation. The point multiplication of the weight and the second-order time difference outputs an enhancement sequence, which retains both mutation information and change sign in amplitude and direction, facilitating the alignment and thresholding processing of the displacement field increment obtained by registration and the fiber micro-strain sequence in time;

[0072] At the parameter level, the slope and the central threshold dominate the sensitivity and the alarm level, the small positive number stabilizes the numerical calculation, and the robust scale from the MAD is insensitive to outliers, which is suitable for complex working conditions such as dust and vibration. Overall, this step constitutes a closed loop from detection to calculation to weight adjustment, providing stable and usable mutation feature input for fusion decision. For example, the sliding window refers to a symmetric or semi-open interval index set around the current sampling point, and the morphological close-open processing is performed with a fixed sample length structure element. The time continuity constraint is realized in a first-order recursive manner and ensures smooth transition of the weight within the event segment. Further, the default window span is 3-10 min and adapts to the sampling period, the morphological structure element has a default length of 3-7 sampling points, the minimum event duration is 2-5 min, and the update period is consistent with the inference step. The above values are based on the statistical distribution of the labeled mutation segments and the false alarm control target. Regarding the interpretation of the formula parameters, the mapping slope coefficient and the central threshold are selected by the ROC optimal point of the validation set or the fixed false positive rate principle, the small positive number is given an order of magnitude based on the sensor quantization bits and the full scale range, and the robust noise scale is estimated by the median absolute deviation of the first-order difference within the window and corrected by quantile when the event duty cycle is high. Optionally, the monotonic S-shaped mapping can be replaced by a piecewise linear approximation when resources are limited, without changing the weight normalization property. The morphological order can also be switched between close-open and open-close according to the false alarm preference. Necessary exception and boundary processing includes: when the window is at the edge of the sequence or there are scattered missing values, mirror filling or nearest neighbor retention is used; when the robust scale estimate approaches zero, an alternative scale based on quantile difference is enabled and the upper limit of the weight change rate is limited.

[0073] In one embodiment, the joint main direction is determined by the principal plane set obtained by clustering the point cloud normals and the strike / dip statistics of the structural surface mapping in the field, and the RQD is used to adjust the normal constraint threshold or the constraint weight.

[0074] In one embodiment, the non-rigid registration adopts a cooperative point drift or a non-rigid method based on a Gaussian mixture model, and adds a normal elastic threshold term and a tangential Coulomb friction term to the objective function to constrain the normal deformation and tangential slip of the joint surface.

[0075] Specifically, in step S5, in the process of registering and solving the multi-period point cloud, GMM / CPD is used as the data consistency backbone, and physical constraints are imposed on the displacement field in the joint main direction: the normal component is limited by the elastic threshold, and the tangential component satisfies the displacement formula equivalent of the Coulomb criterion. The pose and non-rigid deformation coefficient are solved by a unified objective function, a physically interpretable deformation field is obtained, and then output to the subsequent fusion decision; the solving process formula is as follows:

[0076] ,

[0077] wherein, For total energy, For GMM / CPD data items, It is a non-rigid transformation. and Let N and M be the sets of points to be registered and the reference points, respectively, and let N and M be the number of points to be registered and the number of reference points, respectively. This is the GMM variance term, expressed in squares with units of displacement. These are the displacement field smoothness normality coefficients, which are dimensionless. For smoothing operators (such as the Laplace matrix of thin plate splines). The kernel coefficient concatenation vector. These are the normal and tangential constraint weights, respectively. These are the normal and tangential constraint energies, respectively.

[0078] In the formula:

[0079] ,

[0080] ,

[0081] in, For kernel function, Set of control points The first in One control point, To control the number of points, The width of the Gaussian kernel. For the first Displacement coefficient vector of each control point;

[0082] ,

[0083] Among them, the summation index Corresponding points The number, It is a norm 2;

[0084] In the formula: Displacement increment is defined. Define joint set The unit normal is This is obtained from step S4; let each point be mapped to the nearest joint set index. and with distance threshold Forming a constrained index set , The threshold for the point-to-plane distance from a point to its corresponding joint surface, in meters, is determined on-site based on the minimum distance threshold from a point to the joint surface.

[0085] The normal threshold is derived using elastic parameters and monitoring span, and adjusted in conjunction with RQD:

[0086] ,

[0087] ,

[0088] where, is the normal displacement threshold, is the unadjusted reference threshold, is the RQD adjustment coefficient, is the rock mass quality index, 0-100, is the geometric coefficient, dimensionless, is the monitoring span length, is the Poisson’s ratio, is the elastic modulus, is the characteristic normal stress scale, unit Pa;

[0089] The normal constraint energy adopts hinge penalty:

[0090] ,

[0091] where, is the point displacement component in the joint normal direction;

[0092] The tangential component is obtained by the projection matrix :

[0093] ,

[0094] ,

[0095] where, is the 3x3 unit matrix (2x2 if it is a 2D section case), is the tangential slip amount, is the normal compression displacement; the contact stiffness is introduced Displacement of the Coulomb criterion:

[0096] ,

[0097] where, is the friction coefficient of the joint set , dimensionless, which can be set for the joint set, is the cohesion strength, unit Pa, are the normal and tangential contact stiffnesses, respectively, unit Pa / m;

[0098] Weight and solving process:

[0099] 1) Initialization and detection data: use the and RQD, monitoring span calculated in step S4 and set The baseline value is dimensionless and falls within the empirical range. To verify the calibration set;

[0100] 2) Formula calculation: in each outer iteration In the middle, first press the current one. calculate , build The GMM / CPD steps employ closed-form or iterative updates corresponding to the expected values ​​to solve the problem. and ;

[0101] 3) Parameter adjustment: If or If the corresponding violation rate increases, then fine-tuning is performed using a linear strategy. Or appropriately relax / tighten (by (with RQD upper and lower bound constraints), until the residual stabilizes, and the output terminates. With displacement field ;

[0102] Specifically, this step superimposes joint-oriented physical constraints within the probabilistic registration framework, ensuring that the obtained deformation field not only fits the point cloud observations but also follows the stress and structural surface motion characteristics of the rock mass. Data terms are modeled using Gaussian mixtures to establish corresponding relationships, regularization terms restrict violent oscillations in the displacement field, and normal constraints are constructed using elastic modulus, Poisson's ratio, and monitoring span to establish displacement thresholds. These thresholds are then adaptively relaxed or tightened in conjunction with rock mass quality indicators to avoid unreasonable out-of-plane compression or expansion.

[0103] Tangential constraints rewrite the Coulomb friction criterion as a displacement domain inequality, introduce contact stiffness and cohesion, and set the slip tolerance for different joint sets separately. During the optimization process, constraint violations will feed back to the fine adjustment of weights and thresholds, forming a closed loop of observation-computation-parameter adjustment. This design reduces the non-physical deformation that occurs in the fracture-rich region when pure geometric registration occurs. The obtained displacement is decomposed into normal and tangential components, which can be directly called for subsequent fusion and early warning.

[0104] In this embodiment, the control point set is selected on the reference point cloud with voxel down-sampling or uniform grid, the kernel width is matched with the smoothing operator to balance the detail and overall continuity, and the iteration stopping condition is jointly defined by the energy drop threshold and the maximum outer iteration number. Further, the control point spacing is set to 0.5-2.0 m by default, the maximum outer iteration is set to 30-80 times by default, and the energy drop threshold is set to 1e-4-1e-3 by default; these settings are based on the compromise between offline alignment error and convergence speed. Regarding the interpretation of formula parameters, the normal and tangential constraint weights are determined by weighting the loss function of the normal over-limit and tangential misjudgment of the validation set, the elastic modulus and Poisson's ratio in the normal threshold come from indoor tests or in-situ tests, the monitoring span is taken from the construction section scale, the friction coefficient and cohesion come from joint shear tests or inversion, and the normal and tangential contact stiffness is estimated according to the contact model and structure surface roughness. Alternatively, in the case of severe occlusion or severe deformation gradient, multi-resolution registration (coarse first and then fine) or block solution can be used to maintain stable solution and unchanged output interface. Necessary exception and boundary processing includes: when data items and constraint items alternate dominance, temporarily reduce the constraint weight and limit the tightening rate of the normal threshold; if the outer iteration reaches the upper limit and still does not meet the stopping condition, output the current optimal solution and mark it as "registration degradation".

[0105] In one embodiment, the hidden Markov state machine includes three types of hidden states of elasticity-damage evolution-instability, the observation quantity is the amplitude, growth rate and frequency energy of micro-strain, and the state transition probability can be updated online.

[0106] In one embodiment, the hierarchical early warning includes at least three levels, and each level corresponds to different response suggestions and construction control parameter ranges.

[0107] The present application also proposes a tunnel surrounding rock deformation identification system, comprising:

[0108] The multi-source acquisition unit is used for acquiring multi-spectral three-dimensional point cloud, displacement time series and optical fiber micro-strain time series, and completing time synchronization;

[0109] The time series analysis module is used for running the time domain convolution network and the mutation feature strengthening calculation;

[0110] The point cloud processing module is used for executing non-rigid registration based on the Gaussian mixture model with joint constraint and outputting deformation variables;

[0111] The fusion decision module is built-in hidden Markov state machine and physical critical interface, and is used for outputting risk score or critical risk signal;

[0112] The early warning and visualization unit is used for dynamically labeling and linkage displaying the hierarchical early warning results and joint sliding zone in the BIM / construction management model;

[0113] Specifically, the system receives the risk level, the joint main direction and the location attribute of the tangential slip band through a unified data interface of engineering coordinates, and expresses them in the construction management model in terms of section, mileage and arch position. Further, the default data display refresh period is 5 minutes, which can be adjusted to 1-10 minutes, and the single rendering delay does not exceed 30 seconds, which is based on the construction scheduling response requirement. Optionally, when the computing resources are limited, it can be downgraded to only output text and section numbers while omitting the continuous surface rendering, and the linkage logic remains unchanged. When the model interface is unavailable, the system locally caches the latest results and bulk-transmits them after the interface is restored.

[0114] In one embodiment, the physical critical interface receives the yield critical threshold calculated by the Mohr-Coulomb parameters obtained by indoor test or numerical inversion, and supports online updating; similarly, the online updating of the physical critical interface means that the Mohr-Coulomb parameters or inversion results are pulled from the parameter management library at a set period and consistency check is performed, and if the parameter changes exceed the set amplitude, the corresponding threshold is adjusted synchronously and the version number is recorded. Optionally, the update period is 24 hours by default, which can be adjusted to 2-72 hours, and the check rule is determined according to the historical parameter fluctuation range and the upper limit of the test error. When the new parameters do not pass the check or the interface is timed out, the last version of the parameters is used and an internal alarm is triggered to remind manual review.

[0115] In one embodiment, the early warning and visualization unit is configured to: when the high risk level is reached, generate suggestions for grouting reinforcement range and process parameters according to a preset rule library and provide manual confirmation, and highlight the displacement mutation zone and joint slip band in the BIM model;

[0116] For example, the visualization of the displacement mutation zone and the joint slip band uses section highlighting and text labeling consistent with the engineering coordinates, and the color and transparency are mapped according to the risk level, and after clicking, the corresponding time window and key values are displayed. Further, the event browsing and exporting function is retained for 30 days by default, and the export format and fields comply with the data specifications of the construction management system. Optionally, during the night low load period, the system can perform batch playback of historical events for review without changing the online running parameters. When the rendering fails for three consecutive times or the rendering time exceeds the threshold, it is automatically downgraded to only display the section number and risk level text and record the performance event.

[0117] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for part or all of the technical features; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

[0118] Furthermore, to the extent that the terms "comprises", "comprising", "includes", "including" or "has" or "having" are used, these are used as equivalent terms having the same meaning and are intended to be construed as such. Furthermore, the term "comprising" is used in the context of the specification to mean that the specification includes the recited feature, but not to the exclusion of other features. The term "consisting of" is used in the context of the specification to mean that the specification includes only the recited feature, and not other features. The term "consisting essentially of" is used in the context of the specification to mean that the specification includes the recited feature, and other features that do not materially affect the basic operation of the application. Furthermore, to the extent that the term "includes" is used, this term is intended to be equivalent to the term "comprising" and is used in the context to mean that the specification includes the recited feature, but not to the exclusion of other features. Furthermore, the term "substantially" is used in the context of the specification to mean that the recited feature is present in the claimed application, but is not necessarily the only feature present. Furthermore, to the extent that the term "about" is used, this term is intended to be construed as meaning "approximately", "

Claims

1. A method for identifying tunnel soil and rock deformation, characterized in that, include, Step S1: Obtain multi-phase three-dimensional point cloud of tunnel cross section through multispectral laser scanner, obtain settlement / convergence time series data through laser displacement meter, and obtain micro-strain time series data through distributed optical fiber; align the data according to a unified clock, and complete coordinate unification and outlier removal. Step S2: Input the settlement / convergence time series data into a temporal convolutional network to obtain displacement acceleration features including second-order time difference; Step S3: Based on the displacement acceleration amplitude and noise scale estimation within the window, calculate the adaptive enhancement weights. The adaptive enhancement weights are used to weight and enhance the displacement acceleration features to obtain the abrupt feature enhancement sequence. Step S4: Perform normal estimation and plane segmentation on the three-dimensional point cloud, and determine the main direction of the joint surface by combining on-site structural surface mapping / survey line statistics; input rock mass quality indicators, including RQD / Jv, as parameters for constraint strength or prior confidence. Step S5: Non-rigid point cloud registration based on a Gaussian mixture model is used to register and solve the multi-phase 3D point cloud. A joint constraint term is introduced into the energy function to ensure that the deformation field is predominantly tangentially slipping along the principal joint direction, while simultaneously satisfying: The normal deformation does not exceed the threshold determined by the elastic parameters; The tangential sliding amount is limited by the Coulomb friction criterion; Step S6: Input the mutation feature enhancement sequence, the point cloud shape variable obtained based on registration, and the micro-strain time series data into the fusion decision module; Step S7: Analyze the state transition of micro-strain characteristics using a hidden Markov state machine, and output a risk score or critical risk signal. Step S8: Compare the risk score or critical risk signal with the yield critical threshold calculated based on the Mohr-Coulomb parameter, trigger a graded early warning and output the result to the construction management system.

2. The method for identifying tunnel soil and rock deformation as described in claim 1, characterized in that, The multispectral laser scanner simultaneously emits near-infrared and visible light bands and records the echo intensity. Based on the band intensity ratio and spatial neighborhood consistency, it performs dust point discrimination and point cloud completion.

3. The method for identifying tunnel soil and rock deformation as described in claim 1, characterized in that, The adaptive enhancement weights are calculated based on the absolute magnitude of the second-order time difference of the window and a robust noise scale, and time continuity constraints are applied to suppress isolated outliers. The robust noise scale includes the median absolute deviation.

4. The method for identifying tunnel rock and soil deformation as described in claim 1, characterized in that, The main direction of the joint is determined by combining the set of main planes obtained by point cloud normal clustering with the direction / tendency statistics of the on-site structural surface mapping. RQD is used to adjust the normal constraint threshold or constraint weight.

5. The method for identifying tunnel soil and rock deformation as described in claim 1, characterized in that, The non-rigid point cloud registration adopts a cooperative point drift or a non-rigid method based on a Gaussian mixture model, and adds a normal elastic threshold term and a tangential Coulomb friction term to the objective function to constrain the normal deformation and tangential slip of the joint surface.

6. The method for identifying tunnel rock and soil deformation as described in claim 1, characterized in that, The hidden Markov state machine includes three types of hidden states: elastic, damage evolution, and instability. The observables are the amplitude, growth rate, and frequency domain energy of the micro-strain, and the state transition probabilities can be updated online.

7. The method for identifying tunnel rock and soil deformation as described in claim 1, characterized in that, The tiered early warning system includes at least three levels, with each level corresponding to different response recommendations and construction control parameter ranges.

8. A tunnel soil and rock deformation identification system, based on the tunnel soil and rock deformation identification method according to any one of claims 1 to 7, characterized in that, include: The multi-source acquisition unit is used to acquire multispectral three-dimensional point clouds, displacement time series and fiber micro-strain time series, and to complete time synchronization; The time series analysis module is used for runtime domain convolutional network and mutation feature enhancement calculations; The point cloud processing module is used to perform non-rigid registration with joint constraints based on Gaussian mixture models and output deformation variables; The integrated decision module has a built-in hidden Markov state machine and physical critical interface, which is used to output risk scores or critical risk signals. The early warning and visualization unit is used to dynamically annotate and display the graded early warning results and joint sliding zones in the BIM / construction management model.

9. A tunnel rock and soil deformation identification system as described in claim 8, characterized in that, The physical critical interface receives the yield critical threshold calculated from the Mohr-Coulomb parameters obtained from indoor tests or numerical inversion, and supports online updates.

10. A tunnel rock and soil deformation identification system as described in claim 8, characterized in that, The early warning and visualization unit is configured to: when a high risk level is reached, generate suggestions on the grouting reinforcement range and process parameters based on a preset rule base and provide them for manual confirmation, and highlight the displacement abrupt change zone and joint sliding zone in the BIM model.

Images