Reservoir dam body leakage vibration monitoring method based on distributed acoustic sensing

Abstract

The invention discloses a reservoir dam body leakage vibration monitoring method based on distributed sound sensing, and relates to the technical field of dam body safety monitoring, and the method comprises the following steps: obtaining a unified time baseline, constructing a phase reference field, injecting calibration pulses on the basis of the phase reference field, inverting a multi-medium sound velocity tensor, and forming a real-time refraction map; a distributed sensing node arrival angle sequence is obtained under the constraint of a real-time refraction atlas, a wavefront curvature field is solved, an energy anomaly convergence kernel is locked, and a space positioning basis is established for leakage signal feature extraction. According to the invention, a closed loop chain from modeling, purification and positioning to regulation and control is constructed, high-precision leakage identification, dynamic extinguishing of false hot spots and enhancement of real signals are realized through a refraction spectrum, coherent reconstruction and digital twinning, the sensitivity, accuracy and regulation and control capability of dam body leakage monitoring are remarkably improved, and conversion from passive identification to active control is realized.

Description

Technical Field

[0001] This invention relates to the field of dam safety monitoring technology, specifically to a method for monitoring seepage vibration in reservoir dams based on distributed acoustic sensing. Background Technology

[0002] "Reservoir Dam Leakage and Vibration Monitoring Based on Distributed Acoustic Sensing" refers to the real-time sensing and analysis of potential leakage behavior and accompanying weak vibration signals in reservoir dams during operation using a distributed network of acoustic sensors. Leakage causes water flow within the dam body to generate unique acoustic characteristics and vibration patterns through cracks, pores, or joints. These characteristics often precede visible signs of damage. By deploying a large-scale, continuously distributed acoustic sensor network in and around the dam, simultaneous acquisition of sound waves and vibration signals from different locations can be achieved. Combined with signal processing and pattern recognition methods, feature extraction and anomaly detection can be performed, thereby identifying potential leakage hazards and structural risks in the early stages. This method overcomes the limitations of traditional single-point monitoring, offering advantages such as wide coverage, strong real-time performance, and sensitivity to weak anomalies, thus contributing to improved safety monitoring and early warning capabilities for reservoir dams.

[0003] The existing technology has the following shortcomings: In existing technologies, reservoir dam seepage monitoring largely relies on distributed acoustic sensing methods to collect and locate acoustic signals generated by seepage. However, the dam body is typically composed of soil and rock layers with varying densities, concrete layers, and various inclusions. Sound waves propagating at the interfaces of these media are prone to multiple refractions and scattering effects, resulting in highly complex propagation paths. When seepage is in a dynamic evolution phase, localized areas may experience abnormal energy convergence due to refraction effects, manifesting as high-intensity signal hotspots in the sensor network. These false hotspots are easily misidentified as actual seepage locations by existing technologies, obscuring or misidentifying the actual seepage source, directly leading to missed or delayed detection of potential hazards. This locational bias caused by the complex propagation characteristics of the media severely impacts the reliability and timeliness of existing technologies in early warning of dam seepage.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for monitoring seepage vibration in reservoir dams based on distributed acoustic sensing, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring seepage vibration in reservoir dams based on distributed acoustic sensing, comprising the following steps: A unified time baseline is obtained and a phase reference field is constructed. A calibration pulse is injected on the phase reference field to invert the multi-medium sound velocity tensor and form a real-time refraction spectrum, which is used to provide a unified constraint for subsequent signal solution. Under the constraint of real-time refraction patterns, the arrival angle sequence of distributed sensing nodes is obtained, the wavefront curvature field is solved, and the energy anomaly convergence nucleus is located, thus establishing a spatial positioning basis for the feature extraction of leakage signals. Under the influence of energy anomalous convergence nuclei, a percolation-induced multi-frequency sparse dictionary is obtained. By matching local atoms point by point to eliminate non-causal reflection paths, a purified residual spectrum is generated to output a pure signal for subsequent cross-node coherent processing. With the support of purified residual spectrum, cross-node coherent gain constraints are obtained, beam phase rearrangement is performed, and the equivalent direct wave field is reconstructed in the beam phase rearrangement result to form a credible localization evidence chain, providing signal basis for digital twin comparison. Based on a credible chain of location evidence, multi-scale geological parameters are obtained, driving the variable impedance virtual dam body to conduct digital twin comparison, and outputting a refraction deviation correction vector to establish a correction prior for dynamic regulation; Driven by the refraction deviation correction vector, the time reversal control instruction set is acquired, the optical frequency comb phase traction is executed, and the phonon metasurface is driven to inject the phase conjugate acoustic beam, thereby realizing real-time energy transfer traction and dynamic extinguishing of spurious hotspots.

[0007] Preferably, the steps for forming a real-time refractive index are as follows: In the process of obtaining a unified time baseline, multiple acoustic sensors are deployed in different structural areas of the dam. The time consistency between the sensor nodes is constructed by synchronizing the signal with a high-stability time source and loading a delay compensation value. With the support of a unified time baseline, a reference signal is emitted by a controllable sound source, the phase difference of the received signals of each sensing node is collected, and a complete phase reference field is constructed by combining the spatial position relationship of the nodes. Based on the completed phase reference field, high-energy calibration pulses covering multiple frequency bands are injected to collect the arrival time, phase change and amplitude attenuation information of the sensing nodes, and the three-dimensional propagation speed is derived by combining the phase offset characteristics. A real-time refraction spectrum is generated based on the multi-medium sound velocity tensor obtained by inversion, and the spectrum is updated by continuously comparing the propagation parameters to form a dynamic physical reference for constraining subsequent signal solutions.

[0008] Preferably, the energy anomalous convergence core locking process is as follows: Under the constraint of real-time refraction patterns, the signal angle of arrival information of each node is derived based on the positional relationship of multiple acoustic sensing nodes and the calibration impulse response results, combined with the three-dimensional spatial propagation path. Based on the obtained angle of arrival information, a wavefront tangent covering multiple nodes is constructed, the change of normal vector in adjacent regions is calculated, and a wavefront curvature field containing the influence of medium refraction is established. In the wavefront curvature field, the region of curvature extremum clustering is identified, and combined with the energy synchronization enhancement characteristics of multi-node received signals, the three-dimensional coordinate points where energy focusing phenomenon exists are located, and an energy anomaly convergence kernel is constructed. A spatial region of interest is defined with the energy anomaly convergence core as the center. The signals of all sensing nodes in the region are aggregated to construct a local signal cluster, which provides a spatial positioning basis for subsequent leakage signal feature extraction and path inversion.

[0009] Preferably, the steps for generating the purified residual spectrum are as follows: Using the energy anomaly convergence core as a reference, the raw signal data of the surrounding sensing nodes are extracted, and multiple waveform structure parameters are extracted according to frequency to construct a sparse dictionary covering multiple frequency bands. By using a sparse dictionary to match the signals of each node point by point, non-causal reflection paths that are highly consistent with reflective atoms but do not satisfy the propagation causality relationship are identified. The identified non-causal segments are replaced with the corresponding reflective atomic signals in the sparse dictionary. Waveform replacement and difference operations are performed to generate the purified residual spectrum after removing non-causal paths. Using the purified residual spectrum as the input source, time axis alignment and coherence gain calculation are performed on signals from multiple nodes to establish a time-consistent signal network, providing input basis for subsequent propagation direction inversion and localization processing.

[0010] The preferred process for forming a credible chain of evidence for location is as follows: With the support of purified residual spectrum, the principal energy components of the residual spectra of multiple distributed acoustic sensing nodes are extracted and time-aligned. The phase offset matrix is ​​constructed by combining three-dimensional sound velocity tensor data to obtain cross-node coherent gain constraints. Based on coherent gain constraints and phase offset matrix, beam phase rearrangement operation is performed, and subsampling-level time fine-tuning and linear superposition are performed on the residual spectrum signals of each node to generate a joint beam structure with significant spatial focusing effect. The propagation path with the strongest physical consistency is extracted from the joint beam structure, and three-dimensional reverse extension is performed in combination with the sound speed model to reconstruct the equivalent direct wave field, forming a credible positioning evidence chain with causal logic and spatial positioning consistency.

[0011] Preferably, when performing beam phase rearrangement, the sensing node with the strongest phase stability is selected as the main reference source, and the residual spectrum of other nodes is fine-tuned based on its phase characteristics. The fine-tuning includes two steps: time alignment and phase alignment, to ensure that the signals of all nodes achieve maximum energy superposition during beamforming.

[0012] Preferably, the refraction deviation correction vector output process is as follows: Centered on the actual spatial location of the leakage source marked in the credible location evidence chain, we acquire the geophysical parameters covering the location at different spatial scales and construct a multi-scale three-dimensional geological information system. Using the constructed three-dimensional geological information system, a virtual dam body with a real impedance change structure is constructed to drive the digital twin comparison process, simulate the real propagation path of sound waves in each physical propagation unit, and compare it with the measured signal item by item. Based on the propagation deviations identified in the digital twin comparison, a refraction deviation correction vector is constructed by inversion, a dynamically updated correction prior model is established, and it is used to automatically correct the propagation parameters of the consistent region in the subsequent propagation path.

[0013] Preferably, the steps for acquiring the time reversal control instruction set under the drive of the refraction deviation correction vector, executing the optical frequency comb phase pulling, and driving the phonon metasurface to inject the phase conjugate acoustic beam are as follows: Based on the spatial offset, propagation angle deviation, and time delay error information recorded in the refraction deviation correction vector, a set of control instructions with time reversal characteristics is constructed for the reverse reconstruction of the control parameter set of the energy propagation path. Based on the control instruction set, the optical frequency comb phase pulling process is executed, and the constructed control parameters are mapped to each frequency component to generate a multi-frequency composite acoustic excitation source with spatial controllability and time symmetry. A sound wave excitation source is injected into a two-dimensional phonon metasurface. By combining the refraction offset angle and propagation path information in the control command set, a phase-conjugate sound wave beam is output and propagated back along an energy anomaly path. The sound beam is driven into the dam structure, and during the propagation process, it sequentially completes energy de-aggregation, anti-phase interference, and spatial recirculation operations to extinguish false hotspots in the energy focusing area; The wave field structure after regulation is analyzed, and the updated propagation path structure and energy distribution results are fed back to the refraction deviation correction vector library, forming a closed-loop process of monitoring, identification and regulation feedback.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention starts with a unified time baseline and phase reference field, systematically constructs a refraction spectrum, locks down energy anomaly convergence nuclei, and uses sparse dictionary construction and non-causal path elimination techniques to obtain a pure signal spectrum. Combined with beam phase rearrangement and cross-node coherent gain constraints, it achieves high-precision equivalent direct wave reconstruction, thus forming a reliable chain of evidence for seepage location. Based on this, a virtual dam body is constructed by integrating multi-scale geological parameters. A refraction deviation correction vector is formed through digital twin comparison, further driving the injection of optical frequency comb phase traction and phase-conjugate acoustic beams, effectively achieving real-time extinguishing of false hotspots and signal enhancement of real seepage paths. The overall scheme constructs a closed-loop monitoring chain from physical modeling, signal purification, spatial positioning to active control, significantly improving the monitoring system's sensitivity to weak anomalies, positioning accuracy, and interference suppression capabilities. This realizes a shift in early warning of reservoir dam seepage from "passive identification" to "active correction and dynamic control." Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the method for monitoring seepage vibration in reservoir dams based on distributed acoustic sensing, according to the present invention. Detailed Implementation

[0017] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0018] This invention provides, for example Figure 1 The method for monitoring seepage vibration in reservoir dams based on distributed acoustic sensing, as shown, includes the following steps: A unified time baseline is obtained and a phase reference field is constructed. A calibration pulse is injected on the phase reference field to invert the multi-medium sound velocity tensor and form a real-time refraction spectrum, which is used to provide a unified constraint for subsequent signal solution. To address the signal propagation uncertainty caused by the complex internal media of the dam, a complete process is proposed to obtain a unified time baseline and construct a phase reference field. Based on this, calibration pulses are injected to invert the multi-medium sound velocity tensor, forming a real-time refraction spectrum, providing unified constraints for subsequent signal processing. The specific steps are as follows: In obtaining a unified time baseline, multiple acoustic sensors are precisely deployed on the dam surface, at borehole locations within the dam body, and in the contact zone with the dam foundation rock mass, ensuring data collection points at different depths and material interfaces. To maintain time consistency among these sensor nodes, this implementation uses an external high-stability atomic clock to provide a standard time signal. This signal is transmitted via optical fiber to the control terminals around the dam body, and then synchronously transmitted to each sensor node point-by-point via a high-precision distributor. During synchronization, considering the differences in optical fiber transmission delay at different locations within the dam body, the delay time is measured point-by-point, and a compensation value is calculated and applied to the time baseline of each node, ensuring all nodes respond at a unified time scale. This method controls time deviations between nodes within the nanosecond range, maintaining long-term stability even under dynamic changes in dam body temperature and humidity through periodic self-checks and compensation.

[0019] Supported by a unified time baseline, a reference signal with known frequency and amplitude is emitted into the dam body via an externally controllable sound source device. The signal sequentially passes through the dam's concrete overburden, gravel core wall, clay impermeable body, and foundation rock layer. The phase differences of the signal acquired by each sensing node at a unified time scale are completely recorded, forming a multi-point phase data set. In the data processing stage, these phase differences are mapped onto a three-dimensional coordinate system using the spatial relationship between the nodes, thereby constructing a phase reference field covering the entire dam body. This phase reference field accurately reflects the differences in signal propagation speed in different media and can reveal the perturbation law of the propagation phase by the boundaries of each medium. Unlike traditional methods that only use propagation delay as an overall average value, this process comprehensively characterizes the spatial distribution of acoustic properties in different regions.

[0020] After the phase reference field is formed, high-energy calibration pulses are injected into the dam body. These calibration pulses cover multiple frequency bands from low to high frequencies, ensuring clear penetration signals in various media such as gravel, clay, and concrete. The calibration pulses are periodically emitted by an external high-power sound source device, employing different durations and energy levels at different frequency bands to guarantee effective signal propagation even in high-impedance media. Each sensing node receives the calibration pulses and records the specific arrival time, amplitude attenuation, and phase change patterns. These records are then compared point-by-point with the phase reference field constructed in the previous step, and the propagation velocity of each medium in different directions is derived using phase shift and energy change patterns. By combining and analyzing the anisotropic propagation characteristics of all media, a multi-medium sound velocity tensor is gradually obtained through inversion. This tensor not only describes the propagation velocity distribution of sound waves in three-dimensional space but also reveals the directional dependence of each medium, providing a solid foundation for accurately characterizing the acoustic properties of the dam body.

[0021] After the multi-medium sound velocity tensor inversion is completed, the tensor is input into the propagation path calculation process to generate a real-time refraction spectrum covering the entire dam body. The refraction spectrum not only includes the sound velocity distribution but also shows the deflection trajectories of the sound wave propagation path at different incident angles. To ensure the dynamic reliability of the refraction spectrum, the propagation delay and angle of arrival results acquired in real time during monitoring are continuously compared with the refraction spectrum, and deviations in local areas are corrected point by point. The resulting refraction spectrum can adaptively update over time, reflecting in real-time changes in the medium properties within the dam body caused by temperature variations or water seepage. In practical applications, when leakage is in its early stages, the refraction spectrum can immediately show abnormal changes in the local propagation path, thus providing a unified and reliable physical constraint for the accurate calculation of subsequent leakage signals.

[0022] Under the constraint of real-time refraction patterns, the arrival angle sequence of distributed sensing nodes is obtained, the wavefront curvature field is solved, and the energy anomaly convergence nucleus is located, thus establishing a spatial positioning basis for the feature extraction of leakage signals. To achieve accurate signal source localization, based on the existing real-time refraction pattern, the incident direction information of the sensing nodes is further obtained, the spatial distribution of wavefront curvature is constructed, and the energy anomaly convergence nucleus is identified, providing a stable and clear spatial localization basis for leakage signal feature extraction. The specific steps are as follows: Under the constraint of real-time refraction pattern formation, the signal angle of arrival information of each distributed acoustic sensing node is calculated one by one. To achieve this goal, the precise placement of each sensing node in three-dimensional space must first be determined, and this position information must be spatially mapped to the known location of the calibration sound source. During the propagation of the injected calibration pulse, the time information of the acoustic signal received by the sensing node is combined with the sound velocity data in each direction in the real-time refraction pattern to deduce the true propagation direction of each signal path under a specific medium structure. To improve accuracy, the response results of multiple nodes to the same pulse signal are compared. Using the time difference, amplitude change, and phase response data between these nodes, combined with the three-dimensional sound velocity tensor information, a complete incident direction estimation framework is constructed. Through this framework, not only can the specific angle of arrival information of each node for the signal be obtained, but the deflection angle change caused by medium refraction can also be reflected, providing multi-dimensional spatial data support for subsequent wavefront curvature analysis.

[0023] Based on the acquired angle of arrival information from the sensor nodes, the morphology of the entire acoustic wavefront in three-dimensional space is reconstructed, and the wavefront curvature field is calculated accordingly. In this step, multiple spatially distributed nodes that synchronously receive the target signal in time are selected. Using their precise coordinates and the calculated angle of arrival data, the spatial cross-section of the signal wavefront at that time level is reconstructed. By fitting the position and morphology of these wavefront cross-sections in a three-dimensional coordinate system, the rate of change of the normal vector between adjacent regions is calculated to obtain the magnitude of the local wavefront curvature. The construction of the entire wavefront curvature field is based on the gradual superposition and fusion of the local curvature data from all selected nodes, reflecting not only the continuity of the overall wavefront propagation but also revealing its refraction trend when crossing different medium interfaces. To enhance accuracy, the sound velocity variation information provided by the real-time refraction spectrum is incorporated into the local curvature calculation of each node during the curvature field construction process, making the degree of wavefront propagation curvature in different geological regions more consistent with physical reality.

[0024] Based on the reconstructed wavefront curvature field, the anomalous energy convergence nucleus is identified and located, i.e., the location of abrupt changes or anomalous concentrations in wavefront curvature in space. In practice, a numerical scan of each region in the curvature field is first performed to locate curvature extrema or areas of local maxima accumulation. Around each suspected convergence point, the original received signals from multiple nodes are selected, and energy normalization and temporal synchronization are performed to determine whether the region exhibits a physical phenomenon of concentrated propagation of multiple wavefronts. If the region shows a significant focusing trend in the curvature field, and the corresponding signal energy shows synchronous enhancement across multiple nodes, then the point is confirmed as an anomalous energy convergence nucleus in space. The physical significance of such convergence nuclei lies in the fact that sound waves concentrate energy due to multiple refractions or local sound velocity gradients during propagation. This concentration phenomenon often corresponds to potential defect areas within the dam structure, such as loose zones before crack propagation or stress softening zones induced by water seepage.

[0025] After the energy anomaly convergence core is located, a spatial region of interest (ROI) is established centered on its three-dimensional spatial coordinates. This ROI serves as the core reference area for subsequent leakage signal feature extraction and time-domain / frequency-domain analysis. To this end, the target signals received by all distributed sensing nodes within this region are re-aggregated to construct a local signal cluster, and fine-grained resampling is performed to improve time-frequency resolution. Simultaneously, the location of this convergence core is input into subsequent processes such as sparse matching, signal cleansing, and beam reconstruction to determine the initial point and target direction for signal path calculation. When constructing the digital twin model, the coordinates of this convergence core are also used as a real signal reference point for simulation and structural response deduction. Therefore, this spatial ROI is not only the basis for locating the leakage source but also the central benchmark for constructing the entire monitoring, identification, and response logic, possessing significant physical and engineering importance.

[0026] Under the influence of energy anomalous convergence nuclei, a percolation-induced multi-frequency sparse dictionary is obtained. By matching local atoms point by point to eliminate non-causal reflection paths, a purified residual spectrum is generated to output a pure signal for subsequent cross-node coherent processing. To improve signal recognition accuracy, based on the identified energy anomaly convergence nuclei, a seepage-induced multi-frequency sparse dictionary needs to be constructed. This dictionary will then match local signal atoms point-by-point and eliminate non-causal reflection paths, ultimately generating a purified residual spectrum. This provides a highly pure input signal source for subsequent multi-node coherent processing. The specific steps are as follows: Using the spatial center of the energy anomaly convergence nucleus as a reference region, raw leakage-related signal data recorded by all distributed acoustic sensing nodes within this region are extracted to construct a multi-frequency sparse dictionary induced by seepage. During the operation, the raw signals within the target time period are first divided into several continuous frequency bands according to the acoustic frequency from low to high. The frequency division standard covers the entire frequency range from 10 Hz to 1000 Hz to ensure the capture of different spectral characteristics generated by structural responses (such as dam vibration) and fluid-induced vibrations (such as turbulent disturbances in the seepage channel). Within each frequency band, sparse atoms are extracted based on the structural characteristics of the local waveform, specifically including six types of parameters: the rise slope, fall slope, peak amplitude duration, symmetry of both sides of the waveform, time distribution of the dominant frequency abrupt change point, and periodic stability of the signal waveform envelope. Each type of parameter needs to be normalized and stored in a fixed-dimensional vector form, thereby constructing a sparse dictionary spanning frequencies, nodes, and multiple physical feature dimensions. The resulting dictionary is highly recognizable and has clear physical meaning. It specifically reflects the true characteristics of weak vibration signals induced by seepage. Unlike traditional methods that use standardized waveform templates, this dictionary is entirely derived from signal samples inside the target dam body, and has strong adaptability and noise resistance.

[0027] After the sparse dictionary is constructed, signal matching operations need to be performed on each of the sensing nodes around the energy anomaly convergence nucleus to identify potential non-causal reflection paths in the signal. This operation takes the original acoustic signal of a single node as input, extracts its local temporal structural features in segments, and compares each segment of the signal with the atoms in the sparse dictionary one by one. The comparison method not only considers the amplitude similarity of the signal, but also simultaneously analyzes the trend consistency of its waveform structure, the temporal correspondence of abrupt change positions, and the degree of repetition and overlap between the signal periodic structure and the atomic template. If a segment of the signal shows a high degree of matching with multiple sets of reflective atoms in the sparse dictionary, but a low degree of matching with seepage-induced atoms, then that segment is judged to be a non-causal path signal. In the actual environment of the dam, this type of signal mainly originates from multiple reflections and diffractions of sound waves at the interface of the medium, and does not have a unidirectional causal relationship from the source point to the receiver point. It is very easy to cause false leaks and must be eliminated.

[0028] After non-causal path identification is completed, the identification results are used to perform a physical atom substitution and removal operation to form a purified residual spectrum. The specific operation process is as follows: each non-causal segment in the original signal is matched one-to-one with a highly matched reflective atom in the sparse dictionary to construct substitution waveform data. Then, the original waveform of this segment is subtracted from the corresponding atom waveform, and the difference is used as the retained signal. This processing method can preserve the true energy structure in the causal propagation signal to the greatest extent, while removing repetitive echoes and non-physical propagation content caused by propagation path aliasing. Unlike traditional filtering methods, this method does not rely on homogeneous filtering in the frequency domain, but rather performs targeted identification and replacement through physical features and propagation path structure, exhibiting higher selectivity and stronger target retention capability. The final generated residual spectrum presents a continuous time-frequency structure highly correlated with the seepage process, with no spurious repetitions on the time axis and maintaining the true energy profile on the frequency axis, providing a clear and controllable signal foundation for further processing.

[0029] After the purified residual spectrum is generated, it is used as the input source for cross-node coherent processing to establish a high-coherence multi-point signal network. Since the purified residual spectrum has eliminated non-causal components, the residual spectra extracted from multiple sensing nodes will exhibit high synchronization within similar time periods. During the operation, multiple sensing nodes in adjacent or heterogeneous media distribution areas are selected, and their respective purified residual spectra are aligned along the time axis, establishing an equivalent propagation path mapping relationship in the spatial coordinate system. Based on the time-frequency structural similarity and arrival time consistency of the residual spectra, cross-node coherent gain calculation and waveform structure superposition operations are performed. The processing result will be a set of clean signals with high temporal consistency and spatial propagation direction convergence, which can be directly used for subsequent beamforming, wavefront rearrangement, and propagation direction inversion processing.

[0030] With the support of purified residual spectrum, cross-node coherent gain constraints are obtained, beam phase rearrangement is performed, and the equivalent direct wave field is reconstructed in the beam phase rearrangement result to form a credible localization evidence chain, providing signal basis for digital twin comparison. To achieve precise spatial localization of the actual leakage source, based on the generated purification residual spectrum, coherent gain constraints are obtained across nodes, beam phase rearrangement is performed, and the equivalent direct-arrival wavefield is further reconstructed to form a credible chain of physical location evidence, providing traceable physical signal evidence for subsequent digital twin models. The specific steps are as follows: Supported by the clean signal provided by the purified residual spectrum, the temporal relationships, energy structures, and phase evolution trends among nodes in a distributed acoustic sensor network are compared and analyzed to obtain cross-node coherence gain constraints. To achieve this, multiple sensor nodes distributed within the three-dimensional region of the energy anomaly convergence core, having undergone non-causal path removal processing, are selected, and the principal energy components and their temporal peak positions in their residual spectra are extracted. The residual spectra of all nodes are unified to the same time baseline, and the occurrence time of their principal energy peaks and corresponding waveform structures are compared point-by-point. Based on this, a phase offset matrix between nodes is established. This matrix reflects the signal delay and coherence degree caused by differences in propagation paths, geological medium variations, and spatial distances between different nodes. During construction, the three-dimensional sound velocity tensor data provided in the real-time refraction spectrum are used to correct for changes in medium wave velocity and density along the signal propagation path, ensuring that the temporal correction value for each node is consistent with the actual propagation situation. Through this operation, the cooperative propagation capability between each pair of nodes, i.e., the coherence gain index, can be quantified. This coherent gain constraint not only reflects the consistency of signals at each node, but also clarifies which node combinations have the physical basis for joint beam reconstruction, laying a solid data foundation for the next step of phase rearrangement.

[0031] After establishing the cross-node coherent gain constraints, a fine beam phase rearrangement operation is performed on all residual spectrum signals based on the phase offset relationship between the nodes to form a joint wavefield with significant spatial focusing effect. In practice, the group of nodes with the highest cooperative gain is first selected, and the node with the strongest phase stability is used as the master reference source. Then, using this reference node as a benchmark, the residual spectra of all other participating nodes are finely adjusted at the sub-sampling level according to the previously constructed phase offset matrix. This rearrangement operation not only corrects the signal time shift caused by path differences but also further eliminates subtle structural errors between waveforms through refined phase alignment steps, ensuring that the final synthesized beam achieves maximum energy superposition in the main lobe direction. During the adjustment process, wavefront curvature information, the medium type of the beam focusing region, and local sound velocity gradient changes are considered simultaneously, and the contribution weight of each node signal in the spatial mapping is adjusted differentially. After all nodes are adjusted, their waveforms are linearly superimposed to generate a beam structure with a clear propagation direction and concentrated energy. In three-dimensional space, this structure manifests as an energy spine focused towards the actual leakage source, exhibiting significant signal enhancement and sidelobe suppression capabilities. Unlike traditional techniques that rely solely on regular array direction vectors for beamforming, this invention utilizes non-uniformly distributed actual nodes to achieve higher spatial resolution beam rearrangement through the spatial coordination and phase integration characteristics of the waveform itself.

[0032] After beam phase rearrangement, the propagation path with the strongest physical consistency is extracted from its structure to reconstruct the equivalent direct-arrival wavefield, thereby constructing a credible chain of location evidence. During implementation, a three-dimensional energy distribution analysis of the synthesized beam structure is performed to determine the spatial extension trajectory of the beam's main lobe propagation direction and energy center point. Combining the time alignment data and propagation path derivation logic during phase rearrangement, the physical consistency relationship between signal arrival time, propagation direction, and spatial location in the wavefield can be further clarified. Subsequently, using the sound velocity model provided in the real-time refraction spectrum, the energy path is extended backward from the receiver to the signal's starting position, completing the three-dimensional back-projection of the actual leakage source location. This location result not only has signal support from multiple nodes appearing synchronously but also forms a causal chain of location evidence jointly constructed by multiple stages such as coherent enhancement, phase alignment, and path inversion, forming a complete spatial location evidence chain. This evidence chain has clear data sources, traceable propagation paths, and waveform structures with causal logic, providing high-precision physical input for injecting leakage source points, constructing virtual propagation processes, and correcting model parameters in the digital twin dam model.

[0033] Based on a credible chain of location evidence, multi-scale geological parameters are obtained, driving the variable impedance virtual dam body to conduct digital twin comparison, and outputting a refraction deviation correction vector to establish a correction prior for dynamic regulation; To achieve accurate identification and real-time correction of refraction deviations in the seepage propagation path, it is necessary to obtain multi-scale geological parameters of the entire dam body based on a reliable chain of location evidence. This will drive a digital twin comparison of the variable impedance virtual dam body under real-world conditions, and based on this, output a refraction deviation correction vector, establishing a correction prior with dynamic update capabilities. The specific steps are as follows: Centered on the actual spatial location of the seepage source identified in the credible location evidence chain, this study acquires geophysical parameters covering this location at different spatial scales to construct a three-dimensional geological information system capable of reconstructing the actual geological heterogeneity. Specifically, the study first extracts exploration profiles, borehole lithology logs, ultrasonic testing data, and compaction test parameters corresponding to the seepage source area from real observation data. This yields thirteen key geological indicators, including medium density, porosity, water content, shear wave velocity, P-wave velocity, density change rate, shear modulus, viscoelastic hysteresis angle, joint distribution pattern, stress concentration distribution map, temperature gradient data, and hydrological change frequency. During the data collection process, data is extracted in layers at three scales: centimeter, decimeter, and meter, constructing geological datasets with three spatial resolutions: microscopic, mesoscopic, and macroscopic. During the data projection stage, all collected geological property parameter data are uniformly converted to the same three-dimensional spatial coordinate system as the sound field propagation analysis results. Based on spatial interpolation and zoning characteristic analysis methods, the parameters between discontinuous sampling points are continuously calculated, so that data at all scales form a continuous distribution structure of physical properties in the same space, ensuring that they can be accurately referenced in the subsequent twin modeling process.

[0034] Using the acquired multi-scale geological parameter dataset, a virtual dam with a realistic impedance variation structure was constructed to drive a digital twin comparison process. This process simulates the real sound wave propagation path in a controlled environment and compares it item by item with the measured signal. In the implementation, a virtual dam structure diagram was first constructed based on the aforementioned 3D geological data. The geological media in different regions were divided into physical propagation units according to their impedance characteristics, with each unit possessing independent parameters such as sound velocity, density, reflectivity, and refractive index. Then, using the known location and propagation direction of the calibration pulse excitation source as initial conditions, the entire process of sound wave propagation in different units was simulated within the virtual dam. This process not only simulates the refraction path of the signal between interfaces but also records the propagation delay, energy attenuation rate, waveform change amplitude, and dominant frequency drift trend for each propagation path. After the simulation, the signal waveform extracted from the virtual sensing nodes was compared one-to-one with the purified residual spectrum recorded in the actual sensing nodes. The comparison focused on signal arrival time, main peak structure alignment, energy distribution symmetry, and waveform envelope morphology differences, calibrating the spatial and temporal deviations between the virtual and real propagation paths. This comparison method is based on the full structure comparison of waveforms, which is more rigorous than the traditional path correction method that relies on a single arrival time difference. At the same time, it makes full use of the three-dimensional medium structure information to achieve a true mapping from the data space to the physical space.

[0035] Based on the propagation deviations identified during the digital twin comparison process, a refraction deviation correction vector set is constructed and used to establish a dynamically updated correction prior model, serving as the constraint basis for subsequent signal modulation. In this process, for each comparison path, the offset of each refraction node between its starting and ending points is extracted, quantifying its spatial position deviation, wavefront propagation direction angle deviation, and time delay error relative to the ideal path. These three indicators correspond to the three components of the three-dimensional correction vector, and all deviation data are projected back to the real dam coordinate system with millimeter-level spatial resolution. Each refraction deviation correction vector not only reflects the influence of medium disturbances along a certain propagation path but also serves as a high-precision quantitative expression of the impact of macroscopic medium changes on signal propagation. Finally, this refraction deviation vector set is input into the refraction deviation correction prior model, which has the ability to correct new input signals in real time. In subsequent propagation path calculations, whenever a region consistent with the established correction vector is involved, the propagation path assumption will be automatically corrected, and parameters such as time delay, path direction, and wave velocity gradient will be dynamically updated. The prior model will be automatically corrected after each signal processing to adapt to changes in physical properties inside the dam body caused by moisture migration, temperature fluctuations, or structural stress adjustments. This enables the entire monitoring process to have feedback capabilities and realize a closed-loop logic of monitoring, identification, comparison, and correction.

[0036] Driven by the refraction deviation correction vector, the time reversal control instruction set is acquired, the optical frequency comb phase traction is executed, and the phonon metasurface is driven to inject the phase conjugate acoustic beam, thereby realizing real-time energy transfer traction and dynamic extinguishing of spurious hotspots. To achieve real-time control of abnormal energy accumulation induced by leakage, a control command set with time-reversal characteristics needs to be obtained based on the refraction deviation correction vector. Then, a phase-conjugate acoustic beam is injected into the phonon metasurface via phase traction driven by an optical frequency comb to achieve energy path reconstruction and precise extinguishing of spurious hotspots. The specific implementation method is as follows: Based on the spatial offset, propagation angle deviation, and time delay error information recorded in the refraction deviation correction vector, a control instruction set with time-reversal characteristics is constructed. This step analyzes the actual propagation trajectory of each anomalous energy path, extracting the propagation node sequence, medium impedance change locations, abrupt changes in sound velocity gradient, and refraction interface parameters experienced by the signal from the excitation source to the high-intensity energy focusing interval. These parameters are then reconstructed in reverse to form a set of control parameters that satisfy temporal symmetry, propagation direction reversal, and complementary phase structure. This set includes: a propagation path reverse order table, a reverse propagation time table, a wavefront phase reconstruction table, a medium interface sequence index table, and a control window segmentation scale. This instruction set provides a complete physical mapping foundation for subsequent wavefield reconstruction and energy traction, ensuring a high degree of consistency between the reverse propagation logic and the forward path in terms of geometry and physical parameters.

[0037] Based on the control command set, the phase-pulling process of the optical frequency comb is executed to construct a precise excitation source covering the target frequency band. Specifically, a tunable laser array generates a stable optical frequency comb signal, with its frequency interval precisely controlled within the dominant frequency range of the leakage-induced acoustic waves, typically from tens of kilohertz to several megahertz. The phase reversal information from the control command set is mapped onto each comb tooth, constructing a frequency domain structure map with phase reversal characteristics. An electro-optic modulation device converts the optical frequency comb signal into an acoustic wave signal, generating a set of spatially controllable and temporally symmetric multi-frequency composite acoustic wave excitation sources. Each frequency component in this acoustic wave set possesses an anti-phase structure, thus forming a mirror waveform in the time domain, providing a priori driving source for conjugate beam generation.

[0038] The acoustic excitation signal generated by the phase traction of the optical frequency comb is input into a two-dimensional phonon metasurface structure deployed in key areas of the dam. Through programmable cell reflection phase adjustment, spatial wavefront reconstruction and conjugate beamforming are achieved. Each phonon metasurface unit contains an acoustically controlled response material whose acoustic impedance can be adjusted according to an external electrical signal, thereby precisely controlling the reflection phase of the incident wave in each unit. Based on the refraction offset angle and propagation backtracking path information provided in the control instruction set, the control angle and output intensity of each unit are dynamically configured, ensuring that the output wavefield is mirror-symmetrical to the incident wavefield in terms of propagation direction, wavefront morphology, and temporal structure. The output phase-conjugate acoustic beam possesses propagation path backtracking and focus controllability, enabling it to propagate back in the opposite direction of the original energy anomalous focusing path, directly reaching the location where energy accumulation previously occurred, achieving spatial reversal of the wavefield structure.

[0039] A phase-conjugate acoustic beam is injected into the dam structure along an identified energy anomaly path, achieving spatial traction and focusing extinguishing of energy during beam propagation. After entering the dam medium, the acoustic beam undergoes refraction at each medium transition interface, reversing the forward propagation process, and energy redistribution occurs at each refraction interface. Since the beam structure is a complete mirror image of the forward wave field, its energy will defocus in the original focusing region, transforming from a high-intensity focused state to a spatially diffused state. High-energy nodes formed in the forward propagation path are canceled out by the corresponding antiphase waves in the reverse propagation wave field, thus reducing the energy density at that point through physical interference. During this process, energy is redirected to the non-focused region and flows back along the propagation path to the vicinity of the excitation source. Through this process, spurious hotspots caused by the inhomogeneity of the medium structure or the complexity of the propagation path are substantially weakened or even completely extinguished, restoring the acoustic energy distribution throughout the dam to a structural pattern centered on the actual leakage source.

[0040] The wavefield structure after energy traction and spurious hotspot extinction is reanalyzed, and the new propagation path structure, energy distribution spectrum, and control parameters are fed back into the refraction deviation correction vector library to further enrich and update its content. In subsequent monitoring cycles, based on the updated refraction deviation correction vector library, newly emerging energy anomaly regions can be quickly identified and responded to, achieving closed-loop control throughout the entire process from identification, inversion, control, extinction to feedback.

[0041] This invention starts with a unified time baseline and phase reference field, systematically constructs a refraction spectrum, locks down energy anomaly convergence nuclei, and uses sparse dictionary construction and non-causal path elimination techniques to obtain a pure signal spectrum. Combined with beam phase rearrangement and cross-node coherent gain constraints, it achieves high-precision equivalent direct wave reconstruction, thus forming a reliable chain of evidence for seepage location. Based on this, a virtual dam body is constructed by integrating multi-scale geological parameters. A refraction deviation correction vector is formed through digital twin comparison, further driving the injection of optical frequency comb phase traction and phase-conjugate acoustic beams, effectively achieving real-time extinguishing of false hotspots and signal enhancement of real seepage paths. The overall scheme constructs a closed-loop monitoring chain from physical modeling, signal purification, spatial positioning to active control, significantly improving the monitoring system's sensitivity to weak anomalies, positioning accuracy, and interference suppression capabilities. This realizes a shift in early warning of reservoir dam seepage from "passive identification" to "active correction and dynamic control."

[0042] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for monitoring seepage vibration in reservoir dams based on distributed acoustic sensing, characterized in that, Includes the following steps: A unified time baseline is obtained and a phase reference field is constructed. A calibration pulse is injected on the phase reference field to invert the multi-medium sound velocity tensor and form a real-time refraction spectrum. Under the constraint of real-time refraction patterns, the arrival angle sequence of distributed sensing nodes is obtained, the wavefront curvature field is solved, and the energy anomaly convergence nucleus is located, thus establishing a spatial positioning basis for the feature extraction of leakage signals. Under the influence of energy anomalous convergence nuclei, a seepage-induced multi-frequency sparse dictionary is obtained. By matching local atoms point by point to eliminate non-causal reflection paths, a purified residual spectrum is generated. With the support of purified residual spectrum, cross-node coherent gain constraints are obtained, beam phase rearrangement is performed, and the equivalent direct wave field is reconstructed in the beam phase rearrangement result to form a credible localization evidence chain, providing signal basis for digital twin comparison. Based on a credible chain of location evidence, multi-scale geological parameters are obtained, driving the variable impedance virtual dam body to conduct digital twin comparison, and outputting a refraction deviation correction vector; Driven by the refraction deviation correction vector, the time reversal control instruction set is acquired, the optical frequency comb phase traction is executed, and the phonon metasurface is driven to inject the phase conjugate acoustic beam.

2. The method for monitoring seepage vibration of a reservoir dam based on distributed acoustic sensing according to claim 1, characterized in that, The steps for forming a real-time refraction pattern are as follows: In the process of obtaining a unified time baseline, multiple acoustic sensors are deployed in different structural areas of the dam. The time consistency between the sensor nodes is constructed by synchronizing the signal with a high-stability time source and loading a delay compensation value. With the support of a unified time baseline, a reference signal is emitted by a controllable sound source, the phase difference of the received signals of each sensing node is collected, and a complete phase reference field is constructed by combining the spatial position relationship of the nodes. Based on the completed phase reference field, high-energy calibration pulses covering multiple frequency bands are injected to collect the arrival time, phase change and amplitude attenuation information of the sensing nodes, and the three-dimensional propagation speed is derived by combining the phase offset characteristics. A real-time refraction spectrum is generated based on the multi-medium sound velocity tensor obtained by inversion, and the spectrum is updated by continuously comparing the propagation parameters to form a dynamic physical reference for constraining subsequent signal calculations.

3. The method for monitoring seepage vibration of a reservoir dam based on distributed acoustic sensing according to claim 2, characterized in that, The energy anomalous convergence nucleus locking process is as follows: Under the constraint of real-time refraction patterns, the signal angle of arrival information of each node is derived based on the positional relationship of multiple acoustic sensing nodes and the calibration impulse response results, combined with the three-dimensional spatial propagation path. Based on the obtained angle of arrival information, a wavefront tangent covering multiple nodes is constructed, the change of normal vector in adjacent regions is calculated, and a wavefront curvature field is established. In the wavefront curvature field, the region of curvature extremum clustering is identified, and combined with the energy synchronization enhancement characteristics of multi-node received signals, the three-dimensional coordinate points where energy focusing phenomenon exists are located, and an energy anomaly convergence kernel is constructed. A spatial region of interest is defined centered on an energy anomaly convergence core, and signals from all sensing nodes within this region are aggregated to construct a local signal cluster.

4. The method for monitoring seepage vibration of a reservoir dam based on distributed acoustic sensing according to claim 3, characterized in that, The steps for generating the purified residual spectrum are as follows: Using the energy anomaly convergence core as a reference, the raw signal data of the surrounding sensing nodes are extracted, and multiple waveform structure parameters are extracted according to frequency to construct a sparse dictionary covering multiple frequency bands. By using a sparse dictionary to match the signals of each node point by point, non-causal reflection paths that are highly consistent with reflective atoms but do not satisfy the propagation causality relationship are identified. The identified non-causal segments are replaced with the corresponding reflective atomic signals in the sparse dictionary. Waveform replacement and difference operations are performed to generate the purified residual spectrum after removing non-causal paths. Using the purified residual spectrum as the input source, time axis alignment and coherence gain calculation are performed on signals from multiple nodes to establish a time-consistent signal network, providing input basis for subsequent propagation direction inversion and localization processing.

5. The method for monitoring seepage vibration of a reservoir dam based on distributed acoustic sensing according to claim 4, characterized in that, The process of forming a credible chain of evidence for location is as follows: With the support of purified residual spectrum, the principal energy components of the residual spectra of multiple distributed acoustic sensing nodes are extracted and time-aligned. The phase offset matrix is ​​constructed by combining three-dimensional sound velocity tensor data to obtain cross-node coherent gain constraints. Based on coherent gain constraints and phase offset matrix, beam phase rearrangement operation is performed, and subsampling-level time fine-tuning and linear superposition are performed on the residual spectrum signals of each node to generate a joint beam structure with significant spatial focusing effect. The propagation path with the strongest physical consistency is extracted from the joint beam structure, and three-dimensional reverse extension is performed in combination with the sound speed model to reconstruct the equivalent direct wave field, forming a credible positioning evidence chain with causal logic and spatial positioning consistency.

6. The method for monitoring seepage vibration of a reservoir dam based on distributed acoustic sensing according to claim 5, characterized in that, When performing beam phase rearrangement, the sensing node with the strongest phase stability is selected as the main reference source, and the residual spectrum of other nodes is fine-tuned based on its phase characteristics. The fine-tuning includes two steps: time alignment and phase alignment, to ensure that the signals of all nodes achieve maximum energy superposition during beamforming.

7. The method for monitoring seepage vibration of a reservoir dam based on distributed acoustic sensing according to claim 5, characterized in that, The refraction deviation correction vector output process is as follows: Centered on the actual spatial location of the leakage source marked in the credible location evidence chain, we acquire the geophysical parameters covering the location at different spatial scales and construct a multi-scale three-dimensional geological information system. Using the constructed three-dimensional geological information system, a virtual dam body with a real impedance change structure is constructed to drive the digital twin comparison process, simulate the real propagation path of sound waves in each physical propagation unit, and compare it with the measured signal item by item. Based on the propagation deviations identified in the digital twin comparison, a refraction deviation correction vector is constructed by inversion, a dynamically updated correction prior model is established, and it is used to automatically correct the propagation parameters of the consistent region in the subsequent propagation path.

8. The method for monitoring seepage vibration of a reservoir dam based on distributed acoustic sensing according to claim 7, characterized in that, Driven by the refraction deviation correction vector, the time reversal control instruction set is acquired, and the optical frequency comb phase pulling is executed to drive the phonon metasurface to inject the phase conjugate acoustic beam. The steps are as follows: Based on the spatial offset, propagation angle deviation, and time delay error information recorded in the refraction deviation correction vector, a set of control instructions with time reversal characteristics is constructed for the reverse reconstruction of the control parameter set of the energy propagation path. Based on the control instruction set, the optical frequency comb phase pulling process is executed, and the constructed control parameters are mapped to each frequency component to generate a multi-frequency composite acoustic excitation source with spatial controllability and time symmetry. A sound wave excitation source is injected into a two-dimensional phonon metasurface. By combining the refraction offset angle and propagation path information in the control command set, a phase-conjugate sound wave beam is output and propagated back along an energy anomaly path. The sound beam is driven into the dam structure, and during the propagation process, it sequentially completes energy de-aggregation, anti-phase interference, and spatial recirculation operations to extinguish false hotspots in the energy focusing area; The wave field structure after regulation is analyzed, and the updated propagation path structure and energy distribution results are fed back to the refraction deviation correction vector library, forming a closed-loop process of monitoring, identification and regulation feedback.

Images