Excavation-following seismic wave AI imaging system based on seismic signal real-time processing

Abstract

The invention relates to the technical field of tunnel and underground engineering geological exploration, in particular to a seismic signal real-time processing-based digging seismic wave AI imaging system, which comprises a cutterhead vibration coherent mode extraction module, an AI surrounding rock wave velocity structure inversion module, a wave velocity model optimization module and a dynamic geological interface imaging module, wherein the cutterhead vibration coherent mode extraction module is used for extracting a dominant coherent mode; the AI surrounding rock wave velocity structure inversion module is used for obtaining a wave velocity structure prediction result in a pre-trained AI surrounding rock wave velocity inversion network; and the wave velocity model optimization module is used for carrying out residual error correction to obtain an optimized wave velocity structure model. According to the invention, through organic combination of cutterhead vibration sensing, surrounding rock wave velocity intelligent inversion, travel time residual error correction and excavation-following time-shifting imaging, real-time updating and spatial synchronization of a geological interface imaging result in the excavation process are realized, so that the continuous sensing capability of excavation-following earthquake imaging on the change of a front geological structure is improved.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology for tunnels and underground engineering, and in particular to an AI imaging system for seismic waves during tunneling based on real-time processing of seismic signals. Background Technology

[0002] During tunnel and underground engineering construction, the spatial distribution of adverse geological bodies such as surrounding rock structures, faults, fracture zones, and weak interlayers directly affects tunneling safety and construction efficiency. To understand the geological conditions ahead of the tunnel face in advance, seismic wave detection and other methods are commonly used to predict and image the surrounding rock ahead. However, due to the strong non-stationarity of seismic signals under tunneling conditions, and the combined effects of rock-breaking vibrations from the tunnel boring machine, construction noise, and the heterogeneity of the surrounding rock, traditional imaging methods that rely on fixed seismic sources or static wave velocity models are difficult to keep in line with the actual construction conditions. This results in delayed updates of imaging results and accumulation of spatial alignment errors, making it difficult to meet the real-time sensing requirements during continuous tunneling.

[0003] Existing seismic imaging technologies used in tunneling generally suffer from the following shortcomings: First, they fail to effectively utilize the stable excitation source of cutterhead rock-breaking vibration, leading to a disconnect between the source conditions and the construction status. Second, the surrounding rock wave velocity structure largely relies on offline inversion or empirical settings, making it difficult to reflect the dynamic changes in medium conditions during tunneling. Third, the lack of a time-shift correction mechanism synchronized with tunneling displacement during imaging results easily leads to offsets between the imaging results and the actual spatial location, affecting the reliability of geological interface identification. Therefore, it is necessary to propose a seismic wave AI imaging system based on real-time seismic signal processing to address the aforementioned technical challenges. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides an AI imaging system for seismic waves during tunneling based on real-time processing of seismic signals.

[0005] A seismic wave AI imaging system based on real-time seismic signal processing during excavation includes a cutterhead vibration coherence mode extraction module, an AI surrounding rock wave velocity structure inversion module, a wave velocity model optimization module, and a dynamic geological interface imaging module; wherein:

[0006] The cutterhead vibration coherence mode extraction module is used to collect the rock-breaking vibration signal of the cutterhead in real time through a sensor array arranged at equal intervals along the circumference of the tunneling machine cutterhead. Based on the phase difference of the sensor signals, a spatial coherence matrix of the cutterhead vibration is constructed. Eigenvalue decomposition is performed on the spatial coherence matrix to extract the dominant coherence mode. Based on the dominant coherence mode, the equivalent impedance distribution of the contact interface between the cutterhead and the surrounding rock is calculated, and the equivalent impedance distribution is used as a parameter to describe the geological state of the tunnel face.

[0007] AI surrounding rock wave velocity structure inversion module: used to input the geological state description parameters of the tunnel face into the pre-trained AI surrounding rock wave velocity inversion network to obtain the wave velocity structure prediction results of the surrounding rock in front of the tunnel face;

[0008] Wave velocity model optimization module: Based on the seismic wave signals collected by the receiving sensor array deployed on the tunnel sidewall, the first arrival wave travel time information is extracted, and the wave velocity structure prediction results output by the AI ​​surrounding rock wave velocity structure inversion module are used to perform residual correction on the first arrival wave travel time information, thereby obtaining an optimized wave velocity structure model.

[0009] Dynamic geological interface imaging module: Based on the optimized wave velocity structure model, a dynamic imaging operator is constructed, and the seismic wave signal is backpropagated using the dynamic imaging operator to obtain the spatial location distribution of the geological interface; combined with the real-time advance speed of the tunneling machine, the spatial location distribution is time-shifted to generate geological imaging results that are dynamically updated with tunneling.

[0010] Optionally, the cutterhead vibration coherent mode extraction module includes a vibration signal acquisition unit, a phase difference calculation unit, a spatial coherence matrix construction unit, a coherent mode decomposition unit, and an equivalent impedance calculation unit; wherein:

[0011] Vibration signal acquisition unit: It is used to arrange multiple vibration sensors around the cutterhead of the tunneling machine at preset angle intervals, and synchronously acquire the radial vibration signals generated by the cutterhead breaking rock during the tunneling process. Based on a unified time reference, the vibration signals output by each sensor are time-aligned to form a multi-channel vibration signal sequence.

[0012] Phase difference calculation unit: used to perform frequency domain transformation on multi-channel vibration signal sequences within the same time window, extract the instantaneous phase value of each channel in the target frequency band, calculate the phase difference between vibration signals of any two sensors, and generate a phase difference sequence characterizing the phase relationship of the circumferential vibration of the cutterhead;

[0013] Spatial coherence matrix construction unit: Constructs a spatial coherence matrix of the cutterhead vibration based on the phase difference sequence. The matrix elements of the spatial coherence matrix are obtained by calculating the phase difference correlation of the corresponding sensor pairs.

[0014] Coherence mode decomposition unit: used to perform eigenvalue decomposition operation on the spatial coherence matrix to obtain the corresponding eigenvalue set and eigenvector set, and select the eigenvector with the highest energy proportion according to the eigenvalue size as the dominant coherence mode, which is used to characterize the main spatial coherence features in the rock breaking vibration of the cutterhead;

[0015] Equivalent impedance calculation unit: Based on the vibration amplitude and phase distribution relationship of each sensor channel in the dominant coherent mode, combined with the cutterhead structural parameters, the equivalent impedance distribution of the contact interface between the cutterhead and the surrounding rock in the circumferential direction is calculated, and the equivalent impedance distribution is output as a geological state description parameter of the face to the AI ​​surrounding rock wave velocity structure inversion module.

[0016] Optionally, the phase difference calculation unit includes:

[0017] Time window segmentation subunit: used to segment the multi-channel vibration signal sequence according to a preset window length and a preset step size to obtain multi-channel vibration data segments within the same time window, and to apply the same window function to each channel data segment to form a time window signal set for frequency domain analysis;

[0018] Frequency domain transform subunit: used to perform discrete Fourier transform on each channel signal in the time window signal set to obtain the complex spectrum representation of the corresponding channel in the time window, and to map the complex spectrum representation of each channel to the same frequency index set to form a multi-channel complex spectrum set;

[0019] Target frequency band phase extraction subunit: used to select a set of frequency indices located within the target frequency band from the multi-channel complex spectrum set, and calculate the instantaneous phase value of each channel within the target frequency band, wherein the instantaneous phase value is the phase angle of the corresponding complex spectrum value;

[0020] Phase difference sequence generation subunit: used for generating channels corresponding to any two sensors. With channel The phase difference is calculated by indexing frequency by frequency within the target frequency band, and the phase differences within the target frequency band are aggregated to generate a phase difference sequence that characterizes the phase relationship of the circumferential vibration of the cutterhead.

[0021] Optionally, the spatial coherence matrix construction unit includes:

[0022] Phase difference alignment subunit: Used to receive phase difference sequences for each pair of sensor channels. With channel The corresponding aggregated phase difference values ​​within the same time window are arranged in a consistent manner according to the sensor pair index, forming a phase difference aligned dataset indexed by sensor pairs;

[0023] Coherence kernel mapping subunit: used to perform coherence kernel mapping operation on the phase difference value of each sensor pair in the phase difference alignment dataset, mapping the phase difference value to a coherence coefficient that characterizes phase consistency, so as to obtain the set of coherence coefficients corresponding to each sensor pair;

[0024] Matrix element statistics subunit: used to construct a spatial coherence matrix according to the sensor index, taking the coherence coefficient set as input, and then... Line number The matrix elements of the column are set as channels. With channel The coherence coefficient is calculated, and the diagonal elements are set to 1 to form the spatial coherence matrix of the cutterhead vibration.

[0025] Matrix normalization sub-unit: used to perform symmetry-normalization on spatial coherence matrices, making And apply magnitude constraints to the matrix elements so that all off-diagonal elements fall into the range. Within the range of values, a standardized spatial coherence matrix is ​​obtained.

[0026] Optionally, the equivalent impedance calculation unit includes:

[0027] Dominant mode amplitude and phase extraction subunit: Used to receive the dominant coherent mode output by the coherent mode decomposition unit, obtain the relative amplitude coefficient and relative phase coefficient of each sensor channel in the dominant coherent mode according to the feature vector corresponding to the dominant coherent mode, and combine the relative amplitude coefficient and relative phase coefficient to form a circumferential amplitude and phase distribution sequence.

[0028] Vibration velocity estimation subunit: Used to receive the multi-channel vibration signal sequence output by the vibration signal acquisition unit, calculate the vibration velocity amplitude for each channel in the target frequency band corresponding to the dominant coherence mode, and obtain the circumferential vibration velocity amplitude sequence.

[0029] Equivalent force mapping sub-unit: used to establish the mapping relationship between the channel amplitude phase and the contact interface equivalent force based on the circumferential amplitude phase distribution sequence and the cutter head structural parameters. It couples the relative amplitude coefficient of each channel with the vibration velocity amplitude and converts it into the equivalent force amplitude of the corresponding circumferential position to obtain the circumferential equivalent force sequence.

[0030] Impedance ratio calculation subunit: used to calculate the equivalent impedance value for each circumferential position. The equivalent impedance value is determined by the ratio of the equivalent force amplitude to the vibration velocity amplitude at the corresponding position, thereby obtaining the circumferential equivalent impedance sequence.

[0031] Circumferential distribution forming sub-unit: used to arrange the circumferential equivalent impedance sequence according to the sensor circumferential angle index, and to interpolate and fill in the missing angle index positions to form the circumferential equivalent impedance distribution of the cutterhead-surround rock contact interface.

[0032] Optionally, the AI ​​surrounding rock wave velocity structure inversion module includes a sample construction unit, a model training unit, and a wave velocity structure inference unit; wherein:

[0033] Sample construction unit: used to obtain the geological state description parameters of the tunnel face output by the cutterhead vibration coherent mode extraction module during historical tunneling, and combine them with the surrounding rock wave velocity structure results obtained by actual measurement in the corresponding time period to construct training sample pairs that correspond one-to-one with the geological state description parameters and the surrounding rock wave velocity structure, forming a sample dataset for network training.

[0034] Model training unit: Used to perform supervised learning training on the preset AI surrounding rock wave velocity inversion network with sample dataset as input. By minimizing the error function between the network output wave velocity structure and the real wave velocity structure in the sample, the network parameters are iteratively updated until the preset convergence condition is met, and the trained AI surrounding rock wave velocity inversion network is obtained.

[0035] Wave velocity structure inference unit: During actual tunneling, the real-time acquired geological state description parameters of the tunnel face are input into the trained AI surrounding rock wave velocity inversion network to perform forward inference calculations and output the predicted results of the surrounding rock wave velocity structure in the spatial range in front of the tunnel face.

[0036] Optionally, the wave velocity model optimization module includes a seismic signal receiving unit, a first arrival wave travel time extraction unit, a predicted travel time calculation unit, a travel time residual correction unit, and a wave velocity model updating unit; wherein:

[0037] Seismic signal receiving unit: It is used to synchronously acquire seismic wave signals generated during the tunneling process through a receiving sensor array deployed on the tunnel sidewall, and to time-align the seismic wave signals of each receiving channel based on a unified timestamp to form a multi-channel received seismic wave signal sequence.

[0038] First arrival wave travel time extraction unit: used to perform energy detection on the multi-channel received seismic wave signal sequence within a preset time window, identify the arrival time of the first time that the seismic wave signal in each channel exceeds the preset energy threshold, and determine the arrival time as the first arrival wave travel time of the corresponding channel, and form a set of first arrival wave travel times;

[0039] Predicted travel time calculation unit: Receives the wave velocity structure prediction results output by the AI ​​surrounding rock wave velocity structure inversion module, combines the spatial geometric relationship between the emission point and each receiving sensor, calculates the predicted first arrival wave travel time under the theoretical propagation path, and forms a predicted travel time set;

[0040] Travel time residual correction unit: Used to compare the first arrival travel time set with the predicted travel time set, and calculate the travel time residual for each receiving channel. The predicted wave velocity structure is corrected based on the travel time residuals to obtain the corrected wave velocity distribution;

[0041] Wave speed model update unit: used to perform inversion update processing on the predicted wave speed structure based on the spatial distribution characteristics of the travel time residuals, and generate an optimized wave speed structure model that is consistent with the measured travel time constraints.

[0042] Optionally, the wave velocity model update unit includes:

[0043] Residual feedback subunit: Used to receive the time residual output from the time residual correction unit. and in accordance with the first Each receiving channel corresponds to a propagation path that will Mapped to the wave velocity model spatial grid, forming a spatial distribution constraint field for the travel time residuals;

[0044] Model incremental solver sub-element: used to solve for the update increment of wave velocity structure under spatially distributed constraint fields. This minimizes the sum of squared residuals between the updated predicted travel time and the measured travel time.

[0045] The wave velocity update subunit is used to superimpose the update increment onto the predicted wave velocity structure to obtain the optimized wave velocity structure model, expressed as: In the formula, To optimize the wave velocity structure model; To predict wave velocity structure; Update increments for wave speed; Spatial location coordinates;

[0046] Convergence Judgment Subunit: Used to determine whether the update has converged based on whether the residual between the updated predicted travel time and the measured travel time is lower than a preset threshold. When the convergence condition is met, the optimized wave velocity structure model is output to the dynamic geological interface imaging module.

[0047] Optionally, the dynamic geological interface imaging module includes an imaging operator construction unit, a backpropagation imaging unit, an interface focusing determination unit, and a time-shift correction output unit; wherein:

[0048] Imaging operator construction unit: Based on the optimized wave velocity structure model, a dynamic imaging operator consistent with the current propagation characteristics of the surrounding rock medium is constructed within the preset imaging space.

[0049] Backpropagation imaging unit: Based on the dynamic imaging operator, backpropagation processing is performed on the seismic wave signals collected by the receiving sensor array on the tunnel sidewall, and the seismic wave signals of each receiving channel are back-projected into the spatial imaging area to obtain the backpropagation wave field reflecting the distribution of underground reflected energy.

[0050] Interface focusing determination unit: used to perform spatial superposition analysis on the backpropagation wave field, identify the location area where the energy focusing degree is significantly enhanced, and determine the spatial location distribution of the geological interface based on the spatial location that meets the focusing criterion;

[0051] Time-shift correction output unit: It is used to receive the real-time advance speed information of the tunnel boring machine, calculate the tunneling displacement based on the time difference between the reference time corresponding to the imaging result and the current tunneling time, and perform time-shift correction on the spatial position distribution of the geological interface along the tunnel axis to generate geological imaging results that are updated synchronously with the tunneling progress.

[0052] Optionally, the dynamic imaging operator includes a wave velocity field constraint part, a spatial discrete propagation part, a time inversion injection part, and a boundary and stability control part.

[0053] The beneficial effects of this invention are:

[0054] This invention directly transforms the tunneling construction state into quantifiable geological state description parameters of the tunnel face by real-time acquisition of cutterhead rock-breaking vibration signals and extraction of stable coherent features during the tunneling process. Based on this, it combines artificial intelligence to invert the surrounding rock wave velocity structure, thereby realizing the dynamic acquisition of surrounding rock propagation characteristics and enabling the seismic imaging process to continuously reflect the real medium state under tunneling conditions.

[0055] This invention introduces a travel-time residual correction and dynamic imaging operator construction mechanism, and combines the tunneling machine's advance speed to perform time-shift correction on the imaging results, so that the geological interface imaging results are updated synchronously with the tunneling footage, avoiding the accumulation of spatial alignment errors, thereby improving the stability and usability of seismic imaging during continuous construction. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a schematic diagram of an AI imaging system according to an embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of the coherent mode extraction module for the cutterhead vibration in an embodiment of the present invention. Detailed Implementation

[0059] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0060] like Figures 1-2 As shown, a seismic wave AI imaging system based on real-time seismic signal processing during excavation includes a cutterhead vibration coherence mode extraction module, an AI surrounding rock wave velocity structure inversion module, a wave velocity model optimization module, and a dynamic geological interface imaging module; wherein:

[0061] The cutterhead vibration coherence mode extraction module is used to collect the rock-breaking vibration signal of the cutterhead in real time through a sensor array arranged at equal intervals along the circumference of the tunneling machine cutterhead. Based on the phase difference of the sensor signals, a spatial coherence matrix of the cutterhead vibration is constructed. Eigenvalue decomposition is performed on the spatial coherence matrix to extract the dominant coherence mode. Based on the dominant coherence mode, the equivalent impedance distribution of the contact interface between the cutterhead and the surrounding rock is calculated, and the equivalent impedance distribution is used as a parameter to describe the geological state of the tunnel face.

[0062] AI surrounding rock wave velocity structure inversion module: used to input the geological state description parameters of the tunnel face into the pre-trained AI surrounding rock wave velocity inversion network to obtain the wave velocity structure prediction results of the surrounding rock in front of the tunnel face;

[0063] Wave velocity model optimization module: Based on the seismic wave signals collected by the receiving sensor array deployed on the tunnel sidewall, the first arrival wave travel time information is extracted, and the wave velocity structure prediction results output by the AI ​​surrounding rock wave velocity structure inversion module are used to perform residual correction on the first arrival wave travel time information, thereby obtaining an optimized wave velocity structure model.

[0064] Dynamic geological interface imaging module: Based on the optimized wave velocity structure model, a dynamic imaging operator is constructed, and the seismic wave signal is backpropagated using the dynamic imaging operator to obtain the spatial location distribution of the geological interface; combined with the real-time advance speed of the tunneling machine, the spatial location distribution is time-shifted to generate geological imaging results that are dynamically updated with tunneling.

[0065] The coherent mode extraction module for cutterhead vibration includes a vibration signal acquisition unit, a phase difference calculation unit, a spatial coherence matrix construction unit, a coherent mode decomposition unit, and an equivalent impedance calculation unit; wherein:

[0066] Vibration signal acquisition unit: It is used to arrange multiple vibration sensors around the cutterhead of the tunneling machine at preset angle intervals, and synchronously acquire the radial vibration signals generated by the cutterhead breaking rock during the tunneling process. Based on a unified time reference, the vibration signals output by each sensor are time-aligned to form a multi-channel vibration signal sequence.

[0067] Phase difference calculation unit: used to perform frequency domain transformation on multi-channel vibration signal sequences within the same time window, extract the instantaneous phase value of each channel in the target frequency band, calculate the phase difference between vibration signals of any two sensors, and generate a phase difference sequence characterizing the phase relationship of the circumferential vibration of the cutterhead;

[0068] Spatial coherence matrix construction unit: The spatial coherence matrix of the cutterhead vibration is constructed based on the phase difference sequence. The matrix elements of the spatial coherence matrix are obtained by calculating the phase difference correlation of the corresponding sensor pairs. Its construction method satisfies the requirements of matrix symmetry and positive definiteness.

[0069] Coherence mode decomposition unit: used to perform eigenvalue decomposition operation on the spatial coherence matrix to obtain the corresponding eigenvalue set and eigenvector set, and select the eigenvector with the highest energy proportion according to the eigenvalue size as the dominant coherence mode, which is used to characterize the main spatial coherence features in the rock breaking vibration of the cutterhead;

[0070] Equivalent Impedance Calculation Unit: Based on the vibration amplitude and phase distribution relationship of each sensor channel in the dominant coherent mode, and combined with the cutterhead structural parameters, the equivalent impedance distribution of the cutterhead-surrounding rock contact interface in the circumferential direction is calculated, and the equivalent impedance distribution is output as a geological state description parameter of the tunnel face to the AI ​​surrounding rock wave velocity structure inversion module. Through the structured cutterhead vibration coherent mode extraction process described above, the above unit can stably extract the dominant spatial coherent information reflecting the interaction characteristics of the cutterhead-surrounding rock during the tunneling process, and convert it into a quantitative equivalent impedance distribution as a geological state description parameter of the tunnel face, providing a clear physical meaning and continuously updatable input basis for subsequent surrounding rock wave velocity structure inversion.

[0071] The phase difference calculation unit includes:

[0072] Time window segmentation subunit: used to segment the multi-channel vibration signal sequence according to a preset window length and a preset step size to obtain multi-channel vibration data segments within the same time window, and to apply the same window function to each channel data segment to form a time window signal set for frequency domain analysis;

[0073] Frequency domain transform subunit: used to perform discrete Fourier transform on each channel signal in the time window signal set to obtain the complex spectrum representation of the corresponding channel in the time window, and to map the complex spectrum representation of each channel to the same frequency index set to form a multi-channel complex spectrum set;

[0074] The expression for the Discrete Fourier Transform is: In the formula, For the first Each channel in frequency index Complex spectrum value at; For the first Each channel at the sampling point The amplitude of the windowed vibration signal at the location; This represents the number of sampling points within the time window. The imaginary unit; Frequency index;

[0075] Target frequency band phase extraction subunit: Used to select a set of frequency indices located within the target frequency band from the multi-channel complex spectrum set, and calculate the instantaneous phase value of each channel within the target frequency band. The instantaneous phase value is the phase angle of the corresponding complex spectrum value; its calculation formula is: In the formula, For the first Each channel in frequency index The instantaneous phase value at that point; Phase angle operation for complex numbers; For the first Each channel in frequency index Complex spectrum value at;

[0076] Phase difference sequence generation subunit: used for generating channels corresponding to any two sensors. With channel The phase difference is calculated frequency-by-frequency indexing within the target frequency band, and the phase differences within the target frequency band are aggregated to generate a phase difference sequence characterizing the phase relationship of the cutterhead's circumferential vibration. The formula for calculating the phase difference is:

[0077] ; In the formula, For channel With channel In frequency index Phase difference at; For channel The instantaneous phase value; For channel The instantaneous phase value; For channel With channel Aggregated phase difference value within the target frequency band; The number of frequency indices participating in aggregation within the target frequency band; the above sub-units perform frequency domain transformation on multi-channel vibration signals within a unified time window and extract instantaneous phases within the target frequency band, thereby calculating and aggregating phase differences for any sensor pair, forming a stable and consistent phase difference sequence, providing an input basis with clear phase relationship constraints for subsequent spatial coherence matrix construction, thereby improving the repeatability and consistency of the dominant coherence mode extraction process.

[0078] The spatial coherence matrix building unit includes:

[0079] Phase difference alignment subunit: Used to receive phase difference sequences for each pair of sensor channels. With channel The corresponding aggregated phase difference values ​​within the same time window are arranged in a consistent manner according to the sensor pair index to form a phase difference aligned dataset indexed by sensor pairs, and to ensure that the phase difference values ​​of each sensor pair correspond to the same target frequency band and the same time window identifier.

[0080] Coherence kernel mapping subunit: This unit performs a coherence kernel mapping operation on the phase difference values ​​of each sensor pair in the phase difference alignment dataset, mapping the phase difference values ​​to coherence coefficients that characterize phase consistency, thus obtaining the set of coherence coefficients for each sensor pair. The expression for the coherence kernel mapping operation is:

[0081] In the formula, For channel With channel The coherence coefficient; For channel With channel The aggregation phase difference value;

[0082] Matrix element statistics subunit: used to construct a spatial coherence matrix according to the sensor index, taking the coherence coefficient set as input, and then... Line number The matrix elements of the column are set as channels. With channel The coherence coefficient is calculated, and the diagonal elements are set to 1 to form the spatial coherence matrix of the cutterhead vibration; the determination of matrix elements is expressed as follows: In the formula, Spatial coherence matrix Matrix elements;

[0083] Matrix normalization sub-unit: used to perform symmetry-normalization on spatial coherence matrices, making And apply magnitude constraints to the matrix elements so that all off-diagonal elements fall into the range. Within the range of values, a standardized spatial coherence matrix for subsequent eigenvalue decomposition is obtained; the above sub-units, by aligning the phase difference sequence, mapping the coherence kernel and organizing it into a matrix, and implementing symmetric consistency and validity constraints on the obtained spatial coherence matrix, can form a structurally stable spatial coherence matrix input that can be directly used for eigenvalue decomposition, thereby providing a consistent matrix basis for the extraction of dominant coherence modes.

[0084] The principle for constructing the spatial coherence matrix of the cutterhead vibration is as follows:

[0085] During tunneling, sensors at different circumferential positions of the cutterhead simultaneously acquire rock-breaking vibration signals. Due to the spatial continuity and correlation of the contact state between the cutterhead and the surrounding rock, the vibration signals acquired by different sensors are not completely independent in phase, but rather exhibit a certain degree of synchronization or relatively stable phase relationship. The principle of constructing the spatial coherence matrix utilizes this phase consistency characteristic: first, the phase difference between any two sensor vibration signals is calculated; then, the phase difference is converted into numerical coherence coefficients through a determined coherence mapping relationship, and organized into a matrix according to the spatial arrangement of the sensors. The resulting matrix can reflect the correlation strength and spatial distribution characteristics between the vibration behaviors at various circumferential positions of the cutterhead as a whole.

[0086] In this scheme, the spatial coherence matrix serves as a key intermediate representation connecting the original vibration signal and subsequent intelligent inversion processing. Its role is to transform multi-channel, time-varying vibration signals into a mathematical description with a clear spatial structure. By performing eigenvalue decomposition on the spatial coherence matrix, the dominant coherence mode can be extracted from the complex vibration information, thereby stably identifying the main spatial characteristics of the cutterhead-surrounding rock interaction. This enables the subsequent calculation of equivalent impedance distribution and inversion of surrounding rock wave velocity structure to be carried out based on structured and noise-resistant input parameters, thus providing a reliable foundation for the characterization and imaging of the geological state during excavation.

[0087] The steps for performing eigenvalue decomposition on the spatial coherence matrix are as follows:

[0088] The normalized spatial coherence matrix output by the spatial coherence matrix construction unit is received, and its dimension is determined to be... ,in The number of vibration sensors arranged circumferentially around the cutter head, and As the input matrix for eigenvalue decomposition operations;

[0089] Eigenvalue decomposition is transformed into the decomposition of feature pairs. The problem to be solved requires that the characteristic equation satisfy the following equation: In the formula, For the first One eigenvalue; To and The corresponding number 1 eigenvector;

[0090] Calculate matrix The characteristic polynomial is given by the following equation: [Equation omitted for brevity], and by solving for its roots, all eigenvalues ​​are obtained. In the formula, For determinant operations; For eigenvalue variables; for The identity matrix; all the roots obtained from the solution constitute the eigenvalue set. ;

[0091] For each eigenvalue Construct a homogeneous linear system of equations, expressed as: In the formula, for The zero vector; by solving the non-zero solutions of the above system of equations, we obtain the vector with... Corresponding feature vector All feature vectors form the feature vector set. ;

[0092] To ensure that different eigenvectors are consistent in numerical scale, normalization is performed on each eigenvector to satisfy the unit norm constraint; the set of normalized eigenvalues ​​is then... With the set of feature vectors Output.

[0093] The equivalent impedance calculation unit includes:

[0094] Dominant mode amplitude and phase extraction subunit: Used to receive the dominant coherent mode output by the coherent mode decomposition unit, obtain the relative amplitude coefficient and relative phase coefficient of each sensor channel in the dominant coherent mode according to the feature vector corresponding to the dominant coherent mode, and combine the relative amplitude coefficient and relative phase coefficient to form a circumferential amplitude and phase distribution sequence.

[0095] Vibration velocity estimation subunit: This unit receives the multi-channel vibration signal sequence output from the vibration signal acquisition unit, calculates the vibration velocity amplitude for each channel within the target frequency band corresponding to the dominant coherent mode, and obtains the circumferential vibration velocity amplitude sequence. The calculation formula is as follows: In the formula, For the first The vibration velocity amplitude of each channel within the target frequency band; The number of frequency indices within the target frequency band; For the first The frequency value corresponding to each frequency index; For the first Each channel in frequency index Complex spectrum value at; The imaginary unit;

[0096] Equivalent force mapping sub-unit: Used to establish the mapping relationship between the channel amplitude phase and the equivalent force at the contact interface based on the circumferential amplitude phase distribution sequence and the cutterhead structural parameters. It couples the relative amplitude coefficient of each channel with the vibration velocity amplitude and converts it into the equivalent force amplitude at the corresponding circumferential position, obtaining the circumferential equivalent force sequence; the conversion formula is: In the formula, For the circumferential direction of the cutterhead The equivalent force amplitude at each location; For the first The relative amplitude coefficients of each channel in the dominant coherent mode; The force-velocity mapping coefficient is determined by the cutter head structural parameters;

[0097] Impedance ratio calculation sub-unit: Used to calculate the equivalent impedance value for each circumferential position. The equivalent impedance value is determined by the ratio of the equivalent force amplitude to the vibration velocity amplitude at the corresponding position, thus obtaining the circumferential equivalent impedance sequence; the expression for the equivalent impedance value is: In the formula, For the circumferential direction of the cutterhead The equivalent impedance value at each location;

[0098] The circumferential distribution forming sub-unit is used to arrange the circumferential equivalent impedance sequence according to the sensor's circumferential angle index and interpolate to complete the missing angle index positions, forming the circumferential equivalent impedance distribution of the contact interface between the cutterhead and the surrounding rock. The above sub-unit extracts the circumferential amplitude and phase characteristics based on the dominant coherent mode, estimates the vibration velocity amplitude in the target frequency band, and establishes the mapping relationship from amplitude and phase to equivalent force. It further calculates the circumferential equivalent impedance using the ratio of force to velocity, thus forming an equivalent impedance distribution with circumferential spatial orientation.

[0099] The AI-based surrounding rock wave velocity structure inversion module includes a sample construction unit, a model training unit, and a wave velocity structure inference unit; among which:

[0100] Sample construction unit: used to obtain the geological state description parameters of the tunnel face output by the cutterhead vibration coherent mode extraction module during historical tunneling, and combine them with the surrounding rock wave velocity structure results obtained by actual measurement in the corresponding time period to construct training sample pairs that correspond one-to-one with the geological state description parameters and the surrounding rock wave velocity structure, forming a sample dataset for network training.

[0101] Model training unit: Used to perform supervised learning training on the preset AI surrounding rock wave velocity inversion network with sample dataset as input. By minimizing the error function between the network output wave velocity structure and the real wave velocity structure in the sample, the network parameters are iteratively updated until the preset convergence condition is met, and the trained AI surrounding rock wave velocity inversion network is obtained.

[0102] Wave velocity structure inference unit: During actual tunneling, it inputs the real-time geological state description parameters of the tunnel face into the trained AI surrounding rock wave velocity inversion network to perform forward inference operations and outputs the predicted results of the surrounding rock wave velocity structure in the spatial range in front of the tunnel face. By establishing a clear correspondence between the geological state description parameters extracted by the cutterhead vibration and the historical surrounding rock wave velocity structure and using them for network training, the AI ​​surrounding rock wave velocity inversion network can quickly output stable wave velocity structure prediction results under real-time tunneling conditions, providing continuous and consistent prior input for subsequent wave velocity model optimization and geological interface imaging.

[0103] The steps for training the AI-based rock velocity inversion network are as follows:

[0104] In historical tunneling data, the input end is the geological state description parameters of the tunnel face output by the cutterhead vibration coherence mode extraction module, aligned by time window; the monitoring end is the true value of the surrounding rock wave velocity structure within the same time period and spatial range; thus forming a one-to-one corresponding sample pair. ,in Parameters describing geological conditions. It is a wave speed structure;

[0105] right and Perform fixed normalization transformations on each sample and save the same set of standardized parameters; construct training batches of data in batches to ensure that the sample dimensions are consistent within each batch.

[0106] Enter batch The input is fed into the AI ​​surrounding rock wave velocity inversion network to obtain the predicted wave velocity structure. ;

[0107] The difference between the predicted wave velocity structure and the true wave velocity structure is used as the training objective, and the mean squared error loss is employed, the expression of which is: In the formula, This is the loss value; This represents the batch sample size. For the first Predicted wave velocity structure for each sample; This corresponds to the true value wave velocity structure; It is a 2-norm;

[0108] Regarding the loss Calculate the gradient of the network parameters, update the network parameters using gradient descent optimization methods, and complete one iteration;

[0109] Training stops when the validation set loss meets the preset threshold and no longer decreases for several consecutive rounds; the network structure and weight parameters are solidified, and the standardized parameters of the input and output are solidified simultaneously to form a trained AI surrounding rock wave velocity inversion network.

[0110] The AI ​​surrounding rock wave velocity inversion network adopts an end-to-end structure of feature encoder + wave velocity decoder;

[0111] Input layer: Input consists of geological condition description parameters of the tunnel face. Its essence is the equivalent impedance distribution and associated characteristic vectors organized according to the circumferential index of the tool head;

[0112] Feature encoder: for Perform multi-level nonlinear mapping to extract implicit characterizations related to changes in the surrounding rock medium. Encoder output As a characteristic feature of the surrounding rock structure;

[0113] Wavespeed decoder: will Mapped to wave velocity structure prediction on the target space grid in front of the tunnel face The output dimension is consistent with the target wave velocity model mesh, ensuring consistency with the wave velocity structure format required for subsequent travel time residual correction.

[0114] The wave velocity model optimization module includes a seismic signal receiving unit, a first arrival wave travel time extraction unit, a predicted travel time calculation unit, a travel time residual correction unit, and a wave velocity model updating unit; wherein:

[0115] Seismic signal receiving unit: It is used to synchronously acquire seismic wave signals generated during the tunneling process through a receiving sensor array deployed on the tunnel sidewall, and to time-align the seismic wave signals of each receiving channel based on a unified timestamp to form a multi-channel received seismic wave signal sequence.

[0116] First arrival wave travel time extraction unit: used to perform energy detection on the multi-channel received seismic wave signal sequence within a preset time window, identify the arrival time of the first time that the seismic wave signal in each channel exceeds the preset energy threshold, and determine the arrival time as the first arrival wave travel time of the corresponding channel, and form a set of first arrival wave travel times;

[0117] Predicted travel time calculation unit: Receives the wave velocity structure prediction results output by the AI ​​surrounding rock wave velocity structure inversion module, combines the spatial geometric relationship between the emission point and each receiving sensor, calculates the predicted first arrival wave travel time under the theoretical propagation path, and forms a predicted travel time set;

[0118] The travel time of the first arrival wave is predicted by integrating the slowness along the propagation path, using the following formula:

[0119] In the formula, For the first Predicted first arrival travel time for each receiving channel; Indicates the distance from the epicenter to the... The theoretical propagation path of a receiving sensor location; Indicates along the path Upper position The predicted wave velocity value at that location is obtained by taking the wave velocity structure value at that location from the output of the AI ​​surrounding rock wave velocity structure inversion module; These are the arc length coordinates along the propagation path, used to identify the position on the path; Let be an arc length infinitesimal element, representing a minimal distance increment along the path;

[0120] The physical meaning of the first arrival wave is the wavefront that arrives at the receiving point first; given the wave velocity and structure, the wave travels a short distance in the medium. The required time is: ,in Known as slowness, it represents the time required to travel a unit distance; the entire propagation path... The sum of the propagation times over each small distance gives the propagation time from the epicenter to the [missing information]. The total propagation time of the receiving sensors, which is also the predicted travel time of the first arrival wave. Therefore, the above formula is essentially an accumulation of propagation time along the path, which is completely consistent with the intuition that the faster the speed, the shorter the travel time; and the longer the path, the longer the travel time, and the dimensions are strictly true: For length, For length / time, therefore For time; by converting the wave velocity structure output by AI into a slowness accumulation calculation along the propagation path, the predicted first arrival wave travel time consistent with the current wave velocity model can be obtained without relying on additional on-site calibration, providing a strict numerical benchmark for subsequent travel time residual calculation and wave velocity model update.

[0121] Travel time residual correction unit: Used to compare the first arrival travel time set with the predicted travel time set, and calculate the travel time residual for each receiving channel. The predicted wave velocity structure is corrected based on the travel time residuals to obtain the corrected wave velocity distribution; the calculation expression for the travel time residuals is as follows: In the formula, For the first The timing residual of each receiving channel; The measured first arrival wave travel time is for the corresponding channel; This refers to the predicted first arrival travel time for the corresponding channel;

[0122] The travel time residual reflects the deviation between the predicted wave velocity structure and the actual medium propagation characteristics. If the predicted first arrival wave travel time on a certain path is too large, it indicates that the overall predicted wave velocity in the corresponding region of that path is too low. Conversely, if the predicted travel time is too small, it indicates that the predicted wave velocity in the corresponding region is too high. Therefore, the travel time residual can serve as direct constraint information for correcting the direction and amplitude of the wave velocity structure, and can be used to make directional adjustments to the predicted wave velocity structure.

[0123] Wave velocity model update unit: This unit performs inversion update processing on the predicted wave velocity structure based on the spatial distribution characteristics of the travel time residuals, generating an optimized wave velocity structure model consistent with the measured travel time constraints. This unit uses the measured first arrival wave travel time to perform residual correction on the AI ​​wave velocity prediction results, so that the wave velocity structure model simultaneously meets the requirements of real-time vibration sensing information and propagation travel time constraints. This results in an optimized wave velocity structure model with higher consistency with the on-site seismic response, providing a reliable propagation medium basis for subsequent dynamic geological interface imaging.

[0124] The wave speed model update unit includes:

[0125] Residual feedback subunit: Used to receive the time residual output from the time residual correction unit. and in accordance with the first Each receiving channel corresponds to a propagation path that will Mapped to the wave velocity model spatial grid, forming a spatial distribution constraint field for the travel time residuals;

[0126] Model incremental solver sub-element: used to solve for the update increment of wave velocity structure under spatially distributed constraint fields. The expression for minimizing the sum of squared residuals between the updated predicted travel time and the measured travel time is: In the formula, Update increments for wave speed; Number of receive channels; For the first The timing residual of each receiving channel;

[0127] In linear inversion based on travel-time residuals, the update increment of wave velocity structure Calculated using the following formula: ,in, For the spatial position of the wave velocity structure Update increments; This represents the travel time residual vector composed of the difference between the measured first arrival wave travel time and the predicted travel time; This represents the sensitivity matrix of travel time to wave speed, where , indicating the first The ray path to the first The sensitivity of wave velocity to individual grid cells; For regularization parameters; The identity matrix is ​​used as the solution process. The essence of this process is to minimize the predicted travel time error while adjusting as few changes in wave velocity as possible, so that the corrected wave velocity structure conforms to the measured travel time without producing unreasonable and drastic wave velocity jumps.

[0128] The wave velocity update subunit is used to superimpose the update increment onto the predicted wave velocity structure to obtain the optimized wave velocity structure model, expressed as: In the formula, To optimize the wave velocity structure model; To predict wave velocity structure; Update increments for wave speed; Spatial location coordinates;

[0129] Convergence Judgment Subunit: This subunit determines whether the update has converged based on whether the residual between the updated predicted travel time and the measured travel time is lower than a preset threshold. When the convergence condition is met, it outputs the optimized wave velocity structure model to the dynamic geological interface imaging module. The above subunit forms spatial constraints by back-projecting the travel time residuals and solves the wave velocity update increment, so that the wave velocity structure can be inverted and updated under the constraints of the measured travel time. This yields an optimized wave velocity structure model that is consistent with the field propagation response, providing a reliable medium parameter basis for the subsequent construction of dynamic imaging operators.

[0130] The dynamic geological interface imaging module comprises an imaging operator construction unit, a backpropagation imaging unit, an interface focusing determination unit, and a time-shift correction output unit; among which:

[0131] Imaging operator construction unit: Based on the optimized wave velocity structure model, a dynamic imaging operator consistent with the current propagation characteristics of the surrounding rock medium is constructed within the preset imaging space. The dynamic imaging operator is used to describe the propagation and time inversion relationship of seismic waves under the current wave velocity conditions.

[0132] Backpropagation imaging unit: Based on the dynamic imaging operator, backpropagation processing is performed on the seismic wave signals collected by the receiving sensor array on the tunnel sidewall, and the seismic wave signals of each receiving channel are back-projected into the spatial imaging area to obtain the backpropagation wave field reflecting the distribution of underground reflected energy.

[0133] Interface focusing determination unit: used to perform spatial superposition analysis on the backpropagation wave field, identify the location area where the energy focusing degree is significantly enhanced, and determine the spatial location distribution of the geological interface based on the spatial location that meets the focusing criterion;

[0134] The time-shift correction output unit receives real-time advance speed information from the tunnel boring machine (TBM), calculates the tunneling displacement based on the time difference between the reference time corresponding to the imaging result and the current tunneling time, and performs time-shift correction on the spatial distribution of the geological interface along the tunnel axis to generate geological imaging results that are updated synchronously with the tunneling progress. This unit constructs a dynamic imaging operator under the constraint of an optimized wave velocity structure model and performs backpropagation imaging on the seismic wave signal, enabling it to obtain the spatial distribution of the geological interface consistent with the current surrounding rock propagation conditions. Combined with the TBM's advance speed for time-shift correction, the imaging results are continuously updated during the tunneling process, thus ensuring the spatiotemporal consistency and sustainable application capability of the geological imaging results during tunneling.

[0135] The steps to identify regions where energy focusing is significantly enhanced include:

[0136] The backpropagation wavefield output by the backpropagation imaging unit is received. A unified imaging time window is determined according to the propagation time corresponding to the received seismic wave signal. The backpropagation wavefield sequence corresponding to each receiving channel is extracted within the time window to form a wavefield dataset for focusing analysis.

[0137] Cross-channel overlay and time-accumulation processing are performed on the wavefield dataset on a spatial grid to converge the backpropagation wavefields from different receiving channels at the same spatial location, generating an image representing the spatial energy distribution; the corresponding formula is:

[0138] In the formula, For spatial location The volume energy value at the location; Number of receive channels; For the first Each receiving channel is located at ,time The amplitude of the backpropagating wave field; The start and end times of the imaging time window;

[0139] The energy values ​​of each spatial grid point in the imaging body are used as the focusing intensity at the corresponding location, and a focusing intensity distribution map is formed throughout the entire imaging space to quantify the degree of convergence of reflected energy at different spatial locations; whereby... In the formula, Spatial location The focus intensity value at that location;

[0140] A local extremum search is performed on the focused intensity distribution map to identify spatial regions where the focused intensity reaches a local maximum and exceeds a preset focused threshold. The extremum regions that satisfy the spatial connectivity constraints are determined as the spatial location distribution of the geological interface. The above steps, by performing cross-channel superposition and energy accumulation of the backpropagation wave field within a unified imaging time window and identifying the location of the geological interface based on the peak region of the focused intensity, can stably characterize the spatial focusing characteristics of reflected energy under multi-channel constraints, thereby providing a reliable spatial positioning basis for geological interface imaging during tunneling.

[0141] How the interface focus determination unit works:

[0142] During backpropagation imaging, seismic wave signals recorded by each receiving sensor are pushed back into the underground space in the opposite direction of propagation. If a geological interface truly exists at a certain spatial location, it will generate a strong reflection signal during forward propagation. During backpropagation, these reflected energies from different receiving channels and at different times will converge simultaneously near that location. The interface focusing determination unit utilizes this physical characteristic of the spatial convergence of multi-channel reflected energy at the real interface location to analyze the backpropagation wave field. Specifically, this unit spatially superimposes and accumulates the energy of the backpropagation wave fields of all receiving channels within a unified imaging time window, thereby obtaining a spatial distribution reflecting the degree of energy concentration. When the energy at certain spatial locations is significantly higher than that of the surrounding area, it indicates that the reflected waves from multiple channels are simultaneously aligned and focused at that location. This phenomenon indicates that the spatial location of that location is highly consistent with the spatial location of the real geological interface. By identifying these areas where energy focusing is significantly enhanced, the spatial distribution of the geological interface can be determined, providing a reliable basis for the dynamic updating of subsequent imaging results during tunneling.

[0143] The time shift correction procedure includes:

[0144] The real-time propulsion speed sequence output by the tunnel boring machine propulsion system is collected, and the propulsion speed sequence is time-aligned with the imaging reference time of the spatial location distribution of the geological interface based on a unified timestamp to form a propulsion speed dataset that can be used for displacement calculation.

[0145] Read the reference time corresponding to the spatial distribution of the geological interface and the current tunneling time, and calculate the time difference between the two. The corresponding expression is: In the formula, For time difference; This is the current moment of tunneling. For imaging reference time;

[0146] Within the time interval corresponding to the time difference, the propulsion speed sequence is integrated and accumulated to obtain the tunneling displacement from the reference time to the current time. The corresponding expression is: In the formula, This refers to the tunneling displacement; For a moment The speed of advancement; For time infinitesimal elements; For imaging reference time; This is the current moment of tunneling;

[0147] The coordinates of each spatial point in the spatial distribution of the geological interface are translated and corrected along the tunnel axis to obtain the corrected set of spatial point coordinates. This achieves time-shift correction of the spatial distribution of the geological interface, and its expression is as follows: In the formula, These are the corrected coordinates of the spatial points; The coordinates of the spatial point before correction; The unit vector along the tunnel axis; This refers to the tunneling displacement;

[0148] The spatial distribution of geological interfaces corresponding to the corrected set of spatial point coordinates is output as a geological imaging result that is updated synchronously with the tunneling progress.

[0149] The dynamic imaging operator consists of four parts: wave velocity field constraint, spatial discrete propagation, time inversion injection, and boundary and stability control.

[0150] The wave velocity field constraint is provided by the optimized wave velocity structure model, which is used to determine the propagation speed at different spatial locations;

[0151] The spatial discrete propagation part transforms the propagation law of continuous medium into a step-by-step progression relationship on a grid, which is used to calculate the evolution of the wave field in the spatial domain;

[0152] The time-reversal injection section is used to inject the received data as a boundary or source term in reverse time, so that the wave field is reversed from the receiver to the possible reflection / interface location.

[0153] The boundary and stability control section is used to constrain the behavior of the wave field at the boundary of the computational domain, avoid numerical reflection and abnormal energy diffusion, and ensure that the imaging calculation can run stably.

[0154] By constructing a time-reversible dynamic imaging operator with an optimized wave velocity structure model as a constraint, the backpropagation imaging process always adopts a propagation rule consistent with the current propagation characteristics of the surrounding rock, thereby ensuring that the geological interface focusing and positioning is based on a matching propagation medium, supporting continuous updating and consistent output of imaging results as tunneling progresses.

[0155] The steps for constructing a dynamic imaging operator are as follows:

[0156] The optimized wave velocity structure model output by the wave velocity model update unit is received, the imaging spatial range, spatial grid resolution and grid coordinate system are determined, and the wave velocity structure model is resampled onto the imaging grid according to the spatial grid to form a wave velocity field for propagation calculation.

[0157] Based on the spatial location of the tunnel's geometric boundary and the receiving sensor array, boundary processing rules for propagation calculation are set, and time axis information corresponding to the received data is configured so that the time step of propagation calculation is consistent with the sampling interval of the seismic wave signal.

[0158] Based on the wave velocity field and time step parameters, a discrete propagation operator is constructed to describe the evolution of the wave field within the spatial grid, enabling it to advance the wave field state at the current moment to the wave field state at the next moment, and ensuring that this advancement relationship holds consistently throughout the entire imaging spatial domain.

[0159] Based on the discrete propagation operator, a time-reverse propagation relationship is constructed to enable the seismic wave signal recorded at the receiver to be injected into the computational domain in reverse time and propagated to the imaging space in the opposite time direction to the forward propagation, thereby satisfying the operator form required for backpropagation imaging.

[0160] The spatial grid, time stepping, boundary treatment, and sensor injection rules of the propagation operator are solidified into a parameter set of the dynamic imaging operator and output to the backpropagation imaging unit through a standardized interface, ensuring that the corresponding dynamic imaging operator is generated with the latest optimized wave velocity structure model in each tunneling update cycle.

[0161] The core of the dynamic imaging operator is to define the propagation law of seismic waves in the surrounding rock using the latest wave velocity structure model, and to solidify this propagation law into a reusable calculation rule. Since the wave velocity structure of the surrounding rock changes with the progress of tunneling and model updates, the imaging operator must be updated accordingly to ensure that the propagation law used during reverse propagation is consistent with the field medium, hence the name dynamic.

[0162] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0163] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A seismic wave AI imaging system based on real-time seismic signal processing during tunneling, characterized in that, It includes a cutterhead vibration coherence mode extraction module, an AI surrounding rock wave velocity structure inversion module, a wave velocity model optimization module, and a dynamic geological interface imaging module; among which: The cutterhead vibration coherence mode extraction module is used to collect the rock-breaking vibration signal of the cutterhead in real time through a sensor array arranged at equal intervals along the circumference of the tunneling machine cutterhead. Based on the phase difference of the sensor signals, a spatial coherence matrix of the cutterhead vibration is constructed. Eigenvalue decomposition is performed on the spatial coherence matrix to extract the dominant coherence mode. Based on the dominant coherence mode, the equivalent impedance distribution of the contact interface between the cutterhead and the surrounding rock is calculated, and the equivalent impedance distribution is used as a parameter to describe the geological state of the tunnel face. AI surrounding rock wave velocity structure inversion module: used to input the geological state description parameters of the tunnel face into the pre-trained AI surrounding rock wave velocity inversion network to obtain the wave velocity structure prediction results of the surrounding rock in front of the tunnel face; Wave velocity model optimization module: Based on the seismic wave signals collected by the receiving sensor array deployed on the tunnel sidewall, the first arrival wave travel time information is extracted, and the wave velocity structure prediction results output by the AI ​​surrounding rock wave velocity structure inversion module are used to perform residual correction on the first arrival wave travel time information, thereby obtaining an optimized wave velocity structure model. Dynamic geological interface imaging module: Based on the optimized wave velocity structure model, a dynamic imaging operator is constructed, and the seismic wave signal is backpropagated using the dynamic imaging operator to obtain the spatial location distribution of the geological interface; combined with the real-time advance speed of the tunneling machine, the spatial location distribution is time-shifted to generate geological imaging results that are dynamically updated with tunneling.

2. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 1, characterized in that, The cutterhead vibration coherent mode extraction module includes a vibration signal acquisition unit, a phase difference calculation unit, a spatial coherence matrix construction unit, a coherent mode decomposition unit, and an equivalent impedance calculation unit; wherein: Vibration signal acquisition unit: It is used to arrange multiple vibration sensors around the cutterhead of the tunneling machine at preset angle intervals, and synchronously acquire the radial vibration signals generated by the cutterhead breaking rock during the tunneling process. Based on a unified time reference, the vibration signals output by each sensor are time-aligned to form a multi-channel vibration signal sequence. Phase difference calculation unit: used to perform frequency domain transformation on multi-channel vibration signal sequences within the same time window, extract the instantaneous phase value of each channel in the target frequency band, calculate the phase difference between vibration signals of any two sensors, and generate a phase difference sequence characterizing the phase relationship of the circumferential vibration of the cutterhead; Spatial coherence matrix construction unit: Constructs a spatial coherence matrix of the cutterhead vibration based on the phase difference sequence. The matrix elements of the spatial coherence matrix are obtained by calculating the phase difference correlation of the corresponding sensor pairs. Coherence mode decomposition unit: used to perform eigenvalue decomposition operation on the spatial coherence matrix to obtain the corresponding eigenvalue set and eigenvector set, and select the eigenvector with the highest energy proportion according to the eigenvalue size as the dominant coherence mode, which is used to characterize the main spatial coherence features in the rock breaking vibration of the cutterhead; Equivalent impedance calculation unit: Based on the vibration amplitude and phase distribution relationship of each sensor channel in the dominant coherent mode, combined with the cutterhead structural parameters, the equivalent impedance distribution of the contact interface between the cutterhead and the surrounding rock in the circumferential direction is calculated, and the equivalent impedance distribution is output as a geological state description parameter of the face to the AI ​​surrounding rock wave velocity structure inversion module.

3. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 2, characterized in that, The phase difference calculation unit includes: Time window segmentation subunit: used to segment the multi-channel vibration signal sequence according to a preset window length and a preset step size to obtain multi-channel vibration data segments within the same time window, and to apply the same window function to each channel data segment to form a time window signal set for frequency domain analysis; Frequency domain transform subunit: used to perform discrete Fourier transform on each channel signal in the time window signal set to obtain the complex spectrum representation of the corresponding channel in the time window, and to map the complex spectrum representation of each channel to the same frequency index set to form a multi-channel complex spectrum set; Target frequency band phase extraction subunit: used to select a set of frequency indices located within the target frequency band from the multi-channel complex spectrum set, and calculate the instantaneous phase value of each channel within the target frequency band, wherein the instantaneous phase value is the phase angle of the corresponding complex spectrum value; Phase difference sequence generation subunit: used for generating channels corresponding to any two sensors. With channel The phase difference is calculated by indexing frequency by frequency within the target frequency band, and the phase differences within the target frequency band are aggregated to generate a phase difference sequence that characterizes the phase relationship of the circumferential vibration of the cutterhead.

4. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 3, characterized in that, The spatial coherence matrix construction unit includes: Phase difference alignment subunit: Used to receive phase difference sequences for each pair of sensor channels. With channel The corresponding aggregated phase difference values ​​within the same time window are arranged in a consistent manner according to the sensor pair index, forming a phase difference aligned dataset indexed by sensor pairs; Coherence kernel mapping subunit: used to perform coherence kernel mapping operation on the phase difference value of each sensor pair in the phase difference alignment dataset, mapping the phase difference value to a coherence coefficient that characterizes phase consistency, so as to obtain the set of coherence coefficients corresponding to each sensor pair; Matrix element statistics subunit: used to construct a spatial coherence matrix according to the sensor index, taking the coherence coefficient set as input, and then... Line 1 The matrix elements of the column are set as channels. With channel The coherence coefficient is calculated, and the diagonal elements are set to 1 to form the spatial coherence matrix of the cutterhead vibration. Matrix normalization sub-unit: used to perform symmetric normalization on spatial coherence matrices, making And apply magnitude constraints to the matrix elements so that all off-diagonal elements fall into the range. Within the range of values, a standardized spatial coherence matrix is ​​obtained.

5. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 4, characterized in that, The equivalent impedance calculation unit includes: Dominant mode amplitude and phase extraction subunit: Used to receive the dominant coherent mode output by the coherent mode decomposition unit, obtain the relative amplitude coefficient and relative phase coefficient of each sensor channel in the dominant coherent mode according to the feature vector corresponding to the dominant coherent mode, and combine the relative amplitude coefficient and relative phase coefficient to form a circumferential amplitude and phase distribution sequence. Vibration velocity estimation subunit: Used to receive the multi-channel vibration signal sequence output by the vibration signal acquisition unit, calculate the vibration velocity amplitude for each channel in the target frequency band corresponding to the dominant coherence mode, and obtain the circumferential vibration velocity amplitude sequence. Equivalent force mapping sub-unit: used to establish the mapping relationship between the channel amplitude phase and the contact interface equivalent force based on the circumferential amplitude phase distribution sequence and the cutter head structural parameters. It couples the relative amplitude coefficient of each channel with the vibration velocity amplitude and converts it into the equivalent force amplitude of the corresponding circumferential position to obtain the circumferential equivalent force sequence. Impedance ratio calculation subunit: used to calculate the equivalent impedance value for each circumferential position. The equivalent impedance value is determined by the ratio of the equivalent force amplitude to the vibration velocity amplitude at the corresponding position, thereby obtaining the circumferential equivalent impedance sequence. Circumferential distribution forming sub-unit: used to arrange the circumferential equivalent impedance sequence according to the sensor circumferential angle index, and to interpolate and fill in the missing angle index positions to form the circumferential equivalent impedance distribution of the cutterhead-surround rock contact interface.

6. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 1, characterized in that, The AI-based surrounding rock wave velocity structure inversion module includes a sample construction unit, a model training unit, and a wave velocity structure inference unit; wherein: Sample construction unit: used to obtain the geological state description parameters of the tunnel face output by the cutterhead vibration coherent mode extraction module during historical tunneling, and combine them with the surrounding rock wave velocity structure results obtained by actual measurement in the corresponding time period to construct training sample pairs that correspond one-to-one with the geological state description parameters and the surrounding rock wave velocity structure, forming a sample dataset for network training. Model training unit: Used to perform supervised learning training on the preset AI surrounding rock wave velocity inversion network with sample dataset as input. By minimizing the error function between the network output wave velocity structure and the real wave velocity structure in the sample, the network parameters are iteratively updated until the preset convergence condition is met, and the trained AI surrounding rock wave velocity inversion network is obtained. Wave velocity structure inference unit: During actual tunneling, the real-time acquired geological state description parameters of the tunnel face are input into the trained AI surrounding rock wave velocity inversion network to perform forward inference calculations and output the predicted results of the surrounding rock wave velocity structure in the spatial range in front of the tunnel face.

7. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 1, characterized in that, The wave velocity model optimization module includes a seismic signal receiving unit, a first arrival wave travel time extraction unit, a predicted travel time calculation unit, a travel time residual correction unit, and a wave velocity model updating unit; wherein: Seismic signal receiving unit: It is used to synchronously acquire seismic wave signals generated during the tunneling process through a receiving sensor array deployed on the tunnel sidewall, and to time-align the seismic wave signals of each receiving channel based on a unified timestamp to form a multi-channel received seismic wave signal sequence. First arrival wave travel time extraction unit: used to perform energy detection on the multi-channel received seismic wave signal sequence within a preset time window, identify the arrival time of the first time that the seismic wave signal in each channel exceeds the preset energy threshold, and determine the arrival time as the first arrival wave travel time of the corresponding channel, and form a set of first arrival wave travel times; Predicted travel time calculation unit: Receives the wave velocity structure prediction results output by the AI ​​surrounding rock wave velocity structure inversion module, combines the spatial geometric relationship between the emission point and each receiving sensor, calculates the predicted first arrival wave travel time under the theoretical propagation path, and forms a predicted travel time set; Travel time residual correction unit: Used to compare the first arrival travel time set with the predicted travel time set, and calculate the travel time residual for each receiving channel. The predicted wave velocity structure is corrected based on the travel time residuals to obtain the corrected wave velocity distribution; Wave speed model update unit: used to perform inversion update processing on the predicted wave speed structure based on the spatial distribution characteristics of the travel time residuals, and generate an optimized wave speed structure model that is consistent with the measured travel time constraints.

8. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 7, characterized in that, The wave velocity model update unit includes: Residual feedback subunit: Used to receive the time residual output from the time residual correction unit. and in accordance with the first Each receiving channel corresponds to a propagation path that will Mapped to the wave velocity model spatial grid, forming a spatial distribution constraint field for the travel time residuals; Model incremental solver sub-element: used to solve for the update increment of wave velocity structure under spatially distributed constraint fields. This minimizes the sum of squared residuals between the updated predicted travel time and the measured travel time. The wave velocity update subunit is used to superimpose the update increment onto the predicted wave velocity structure to obtain the optimized wave velocity structure model, expressed as: In the formula, To optimize the wave velocity structure model; To predict wave velocity structure; Update increments for wave speed; Spatial location coordinates; Convergence Judgment Subunit: Used to determine whether the update has converged based on whether the residual between the updated predicted travel time and the measured travel time is lower than a preset threshold. When the convergence condition is met, the optimized wave velocity structure model is output to the dynamic geological interface imaging module.

9. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 1, characterized in that, The dynamic geological interface imaging module includes an imaging operator construction unit, a backpropagation imaging unit, an interface focusing determination unit, and a time-shift correction output unit; wherein: Imaging operator construction unit: Based on the optimized wave velocity structure model, a dynamic imaging operator consistent with the current propagation characteristics of the surrounding rock medium is constructed within the preset imaging space. Backpropagation imaging unit: Based on the dynamic imaging operator, backpropagation processing is performed on the seismic wave signals collected by the receiving sensor array on the tunnel sidewall, and the seismic wave signals of each receiving channel are back-projected into the spatial imaging area to obtain the backpropagation wave field reflecting the distribution of underground reflected energy. Interface focusing determination unit: used to perform spatial superposition analysis on the backpropagation wave field, identify the location region where the energy focusing degree is significantly enhanced, and determine the spatial location distribution of the geological interface based on the spatial location that meets the focusing criterion; Time-shift correction output unit: It is used to receive the real-time advance speed information of the tunnel boring machine, calculate the tunneling displacement based on the time difference between the reference time corresponding to the imaging result and the current tunneling time, and perform time-shift correction on the spatial position distribution of the geological interface along the tunnel axis to generate geological imaging results that are updated synchronously with the tunneling progress.

10. The AI ​​imaging system for tunneling seismic waves based on real-time seismic signal processing according to claim 9, characterized in that, The dynamic imaging operator includes a wave velocity field constraint part, a spatial discrete propagation part, a time inversion injection part, and a boundary and stability control part.

Images