Mine filling performance detection and stability assessment method and system

Abstract

The present invention belongs to the technical field of mine safety monitoring and geotechnical engineering stability assessment, and relates to a method and system for detecting and assessing the performance of mine filling bodies and their stability. The method comprises the following steps: S1: real-time acquisition of a first type of monitoring signal by a distributed optical fiber sensor array, while simultaneously using a vibration waveform signal as a second type of monitoring signal; S2: inputting the first type of monitoring signal and the second type of monitoring signal into a multi-source data fusion model to generate a fusion state signal that characterizes the three-dimensional damage degree and real-time deformation state of the filling body; S3: inputting the fusion state signal into a parameter dynamic identification module to generate an updated numerical model parameter signal; S4: based on the updated numerical model parameter signal, performing stability quantification calculation to generate a risk level signal. The method and system for detecting and assessing the performance of mine filling bodies of the present invention can solve the problem of inaccurate recognition of psychological states due to large individual differences, strong environmental interference, and weak dynamic adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of mine safety monitoring and geotechnical engineering stability assessment, and in particular to a mine filling body performance detection and stability assessment method and system. Background Art

[0002] Mine backfills are critical structures for underground safety. Their performance degradation and instability can lead to major accidents such as goaf collapse and surface subsidence. Existing technologies rely on laboratory testing to obtain backfill strength parameters, but the sampling process destroys the structure and fails to reflect the evolution of the on-site environment, resulting in significant deviations between the parameters and actual working conditions. Field monitoring primarily uses point sensors (such as strain gauges and displacement gauges), which can only obtain local discrete data and make it difficult to construct an overall three-dimensional state field. While microseismic monitoring can capture internal fractures, it cannot quantify the extent of damage and deformation trends. Geophysical methods (such as resistivity imaging) are limited by the multiplicity of interpretations and low resolution, making them insufficiently practical for engineering applications.

[0003] In the stability assessment process, traditional numerical models use fixed parameters and are unable to track the cumulative damage process of the filling body, resulting in a disconnect between the assessment results and the actual status. The safety warning mechanism based on artificial thresholds ignores the spatiotemporal correlation of damage evolution, resulting in a high false alarm rate and delayed response. The patent with authorization announcement number CN114956749B provides a method for determining the mix ratio of mine filling bodies, but it can only measure the filling ratio value and cannot define its performance and stability. Although some studies have attempted to integrate multi-source data in recent years, they are mostly limited to simple weighted superposition and have not solved the problems of spatiotemporal registration and deep feature extraction of heterogeneous data. Machine learning directly predicts stability, but due to the lack of physical mechanism constraints, it has poor interpretability and high extrapolation risk. The current intelligent construction of mines urgently needs a technical solution that can perceive the three-dimensional damage status of the filling body in real time, dynamically correct the assessment model, and achieve accurate risk grading, so as to break through the technical bottlenecks of data fragmentation, model rigidity, and passive warning. Summary of the Invention

[0004] In view of the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a method and system for detecting and evaluating the performance and stability of mine filling bodies, which is used to solve the problems of one-sided perception of filling body status, inaccurate models, and delayed early warning. The present invention generates a fused state signal that characterizes three-dimensional damage by fusing multi-source heterogeneous signals from distributed optical fibers and microseismic arrays. This signal is used to drive an online inversion algorithm to dynamically update the numerical model parameters, allowing the model to approach the actual state in real time. Based on the updated model, the fused safety factor and failure probability signal are used to generate graded early warning instructions, forming a closed loop of "perception-modeling-assessment", overcoming the shortcomings of traditional methods such as isolated data, static models, and delayed assessment.

[0005] The mine filling performance detection and stability assessment method provided by the present invention includes:

[0006] S1: A distributed optical fiber sensor array is used to collect the spatial distribution signals of strain and temperature in the filling in real time as the first type of monitoring signal. At the same time, a microseismic sensor array is used to collect the vibration waveform signals generated by the internal rupture events of the filling as the second type of monitoring signal.

[0007] S2: Input the first and second monitoring signals into a multi-source data fusion model to generate a fusion status signal representing the three-dimensional damage degree and real-time deformation status of the filling body. The multi-source data fusion model uses a deep learning algorithm to extract features and perform spatial interpolation on heterogeneous temporal and spatial resolution data.

[0008] S3: The fusion state signal is input into the parameter dynamic identification module, and the strength parameters and damage evolution parameters of the filling body are calculated through the online inversion algorithm. This generates an updated numerical model parameter signal and drives the numerical model of the mine filling body to synchronously update its mechanical constitutive relationship.

[0009] S4: Based on the updated numerical model parameter signal, stability quantification calculation is performed to generate a risk level signal. The risk level signal outputs a graded warning instruction by fusing the safety factor output by the numerical simulation and the failure probability predicted by the machine learning model.

[0010] In one embodiment of the present invention, in step S1, a surface displacement field change signal is collected as a third type of monitoring signal by a three-dimensional laser scanning device arranged on the surface of the filling body, and a resistivity distribution signal is collected as a fourth type of monitoring signal by a geoelectric resistivity tomography device buried inside the filling body. In step S2, the third type of monitoring signal and the fourth type of monitoring signal are input into a multi-source data fusion model together with the first type of monitoring signal and the second type of monitoring signal to generate an enhanced fusion status signal that includes both the internal damage status and the surface deformation characteristics.

[0011] In one embodiment of the present invention, the operations performed by the multi-source data fusion model in step S2 include: performing temperature compensation calculation on the first type of monitoring signal to eliminate the interference of ambient temperature on strain measurement, generating a temperature-corrected strain signal, performing waveform feature extraction on the second type of monitoring signal to identify the spatial location and energy level of the rupture event, generating a microseismic event location signal, mapping the temperature-corrected strain signal and the microseismic event location signal to a unified three-dimensional spatial grid through a spatiotemporal alignment algorithm, using a convolutional neural network to perform cross-modal feature fusion on the grid node data, and outputting a fusion status signal that represents the probability of spatial distribution of damage.

[0012] In one embodiment of the present invention, the online inversion algorithm in step S3 adopts an ensemble Kalman filter framework to assimilate the fused state signal into observation data, and iteratively adjusts the numerical model parameters to make the simulated output signal approach the observation signal. When the root mean square error between the simulated strain field signal and the strain component in the fused state signal is less than a set threshold, the strength parameters and damage evolution parameters of the current iteration step are output as the updated numerical model parameter signal.

[0013] In one embodiment of the present invention, the damage evolution parameters include the critical threshold of the damage variable and the damage expansion rate. By real-time tracking the spatiotemporal aggregation characteristics of the microseismic event location signal in the fusion state signal, the attenuation function of the critical threshold of the damage variable is dynamically corrected, so that the updated numerical model can reflect the accelerated effect of the filling body damage accumulation.

[0014] In one embodiment of the present invention, the stability quantification calculation in step S4 includes parallel dual-path processing: the first path inputs the updated numerical model parameter signal into the strength reduction module, iteratively reduces the filling body strength parameter until the numerical model diverges, and outputs a safety factor signal based on limit equilibrium; the second path inputs the fusion state signal into a survival analysis model trained with historical cases, outputs a failure probability signal within a set time period in the future, and finally weightedly fuses the safety factor signal and the failure probability signal to generate a comprehensive risk level signal.

[0015] In one embodiment of the present invention, the generation rule of graded warning instructions is: when the comprehensive risk level signal is in the first interval, a green safety instruction is output; when entering the second interval, a yellow inspection instruction is output and the local high-frequency monitoring mode is triggered; when entering the third interval, an orange warning instruction is output and the emergency reinforcement plan generation module is activated; when entering the fourth interval, a red evacuation instruction is output and the mine dispatching system is linked.

[0016] In one embodiment of the present invention, step S5 is also included: deploying edge computing nodes underground, preprocessing the first type of monitoring signals and the second type of monitoring signals collected in step S1, generating compressed feature signals and uploading them to the cloud, executing the multi-source data fusion model, parameter dynamic identification module and stability quantification calculation in the cloud, and transmitting the generated risk level signal to the edge computing node for conversion into an early warning instruction.

[0017] In one embodiment of the present invention, the edge computing node executes an adaptive monitoring strategy: when the received risk level signal is in the green safety instruction state, the sampling frequency of the first type of monitoring signal is reduced to the basic frequency; when the risk level signal is upgraded to the yellow inspection instruction state, the dormant microseismic sensor array is automatically awakened and a high-precision scanning instruction is triggered.

[0018] The present invention also includes a mine filling performance detection and stability assessment system, including:

[0019] The acquisition module uses a distributed optical fiber sensor array to collect the strain and temperature spatial distribution signals in the filling body in real time as the first type of monitoring signal. At the same time, the microseismic sensor array collects the vibration waveform signal generated by the internal rupture event of the filling body as the second type of monitoring signal.

[0020] A coordination module inputs the first and second monitoring signals into a multi-source data fusion model to generate a fusion state signal representing the three-dimensional damage degree and real-time deformation state of the filling body;

[0021] Verification module: The verification module inputs the fusion state signal into the parameter dynamic identification module, calculates the strength parameters and damage evolution parameters of the filling body through the online inversion algorithm, generates an updated numerical model parameter signal, and drives the numerical model of the mine filling body to synchronously update its mechanical constitutive relationship;

[0022] The preview module performs stability quantification calculation based on the updated numerical model parameter signal to generate a risk level signal. The risk level signal is generated by fusing the safety factor output by the numerical simulation and the failure probability predicted by the machine learning model.

[0023] Beneficial effects: The mine filling performance detection and stability assessment method and system provided by the present invention generate a fusion status signal that characterizes three-dimensional damage by fusing multi-source heterogeneous signals of distributed optical fiber and microseismic array; use this signal to drive the online inversion algorithm to dynamically update the numerical model parameters, so that the model approaches the real state in real time; generate graded warning instructions based on the updated model fusion safety factor and failure probability signal, forming a "perception-modeling-assessment" closed loop, overcoming the defects of traditional methods such as data isolation, static model and delayed assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 This is a system architecture diagram of the mine filling performance detection and stability evaluation system of the present invention;

[0026] Figure 2 The present invention is a flow chart of the method for detecting the performance of mine fillings and evaluating their stability. DETAILED DESCRIPTION

[0027] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0028] See also Figure 1-Figure 2 The mine filling performance detection and stability assessment method of the present invention includes:

[0029] S1: A distributed optical fiber sensor array is used to collect the spatial distribution signals of strain and temperature in the filling in real time as the first type of monitoring signal. At the same time, a microseismic sensor array is used to collect the vibration waveform signals generated by the internal rupture events of the filling as the second type of monitoring signal.

[0030] S2: Input the first and second monitoring signals into a multi-source data fusion model to generate a fusion status signal representing the three-dimensional damage degree and real-time deformation status of the filling body. The multi-source data fusion model uses a deep learning algorithm to extract features and perform spatial interpolation on heterogeneous temporal and spatial resolution data.

[0031] S3: The fusion state signal is input into the parameter dynamic identification module, and the strength parameters and damage evolution parameters of the filling body are calculated through the online inversion algorithm. This generates an updated numerical model parameter signal and drives the numerical model of the mine filling body to synchronously update its mechanical constitutive relationship.

[0032] S4: Based on the updated numerical model parameter signal, stability quantification calculation is performed to generate a risk level signal. The risk level signal outputs a graded warning instruction by fusing the safety factor output by the numerical simulation and the failure probability predicted by the machine learning model.

[0033] like Figure 2As shown, the core process of the mine filling performance detection and stability assessment method begins with the signal acquisition link. The distributed optical fiber sensor array pre-buried in the filling body captures the strain and temperature spatial distribution signals in real time as the first type of monitoring signal. The array is arranged along the key load-bearing path and uses backscattered light analysis technology to obtain axial strain and temperature field data. At the same time, it uses the mining industrial ring network to transmit signals and avoids interference through time division multiplexing. Simultaneously, a microseismic sensor array is deployed on the surface of the surrounding rock in the goaf. The array is distributed according to a three-dimensional grid topology to collect the original signal of the vibration waveform generated by the internal fracture of the filling body. After denoising filtering and event extraction, the second type of monitoring signal containing the fracture location, energy and frequency characteristics is generated. The two types of signals are input into the multi-source data fusion model for deep processing. The model first performs temperature-strain decoupling calculation on the first type of monitoring signal. By separating the strain wavelength drift and the temperature wavelength drift, according to the calibration relationship:

[0034] ;

[0035] in, is the wavelength drift, ε is the true strain, ΔT is the temperature change, and the two coefficients and It needs to be determined through calibration, and the pure temperature signal of the reference optical fiber is used to compensate for the thermal expansion effect and output a temperature-corrected strain signal; at the same time, wavelet denoising and P / S wave separation are performed on the second type of monitoring signal, and the P-wave first arrival time of at least four sensor nodes is selected to establish the travel time equation, and the Geiger iteration method is used to solve the source coordinates and the time of earthquake occurrence, and the magnitude is calculated in combination with the waveform amplitude to generate the microseismic event location signal.

[0036] Furthermore, a three-dimensional Cartesian coordinate system of the filling body was established and divided into equally spaced grids. The temperature-corrected strain signal was mapped to the grid edge according to the optical fiber path, and the node strain value was calculated through cubic spline interpolation. At the same time, the microseismic event location signal was assigned to the corresponding grid according to the source coordinates and the cumulative energy value was counted. A dual-channel input tensor was constructed (channel one stored the grid node strain value, and channel two stored the grid event energy value). The input was input into a three-dimensional convolutional neural network for processing: the first layer of 3×3×3 convolution kernel extracted local strain gradient features, the second layer performed maximum pooling and downsampling, and the third layer of 3×3×3 convolution kernel fused cross-modal features. Finally, the fully connected layer outputted a fusion state signal representing the damage probability of each grid unit. This signal integrated the three-dimensional field data of spatial position, strain value and damage probability in the form of a tensor. In step S1, a surface displacement field change signal is additionally generated as a third type of monitoring signal by a three-dimensional laser scanning device on the surface of the filling body. The device periodically emits a laser point cloud and compares the coordinate changes to calculate the displacement vector field. At the same time, a DC current is injected into the formation using a buried georesistivity tomography electrode array. The pore water distribution and the resistivity spatial distribution of the fracture development zone are reconstructed through potential difference measurement and inversion algorithm to generate a fourth type of monitoring signal. Two new types of signals are involved in multi-source data fusion: the third type of signal extracts the normal displacement component through point cloud registration and resamples it to the three-dimensional grid surface node; the fourth type of signal is converted into damage indication parameters through the resistivity-damage association model and interpolated to fill the internal nodes of the grid. Finally, the input dimension is expanded to a graph neural network. By constructing a graph structure with grid nodes as vertices and spatial adjacency relationships as edges, the graph convolution layer is used to aggregate strain, microseismicity, displacement and resistivity features, output an enhanced fusion status signal and add displacement vector and resistivity change rate fields. After the fusion state signal is input into the parameter dynamic identification module, the ensemble Kalman filter process is started: the strength parameter set and damage evolution parameter set of the numerical model are defined as the state vector signal, the strain component of the fusion state signal is extracted to construct the observation vector signal; the prior distribution is set based on the laboratory test value and the initial parameter set is generated through Latin hypercube sampling; each ensemble member is input into the numerical model in parallel to run the mechanical calculation and output the simulated strain field signal set; the parameter mean vector, covariance matrix and the observed variable covariance matrix and parameter-observed variable cross-covariance matrix of the simulated strain field are calculated; the difference between the actual observation vector and the simulated observation set is multiplied by the Kalman gain matrix to generate the parameter increment adjustment signal (the Kalman gain is determined by the product of the cross-covariance matrix and the inverse of the observation covariance matrix); the root mean square error between the simulated strain field and the actual signal of the updated parameter set is iteratively calculated, and when the error change rate for three consecutive iterations is lower than the convergence threshold, the optimized numerical model parameter signal is output, which is written to the numerical model database in real time through the application program interface to complete the constitutive relationship update.In the in-depth implementation, a dynamic correction mechanism for damage evolution parameters is activated: the cumulative energy value per unit time within the grid is calculated based on the microseismic event location signal of the fused state signal, and converted into a damage development degree indicator through the energy-damage mapping function; when the growth rate of the damage indicator of adjacent grid cells exceeds the preset acceleration threshold, a damage acceleration area coordinate signal is generated and the threshold attenuation function reconstruction is triggered - replacing the standard linear attenuation with an exponential function:.

[0037] ;

[0038] in, is the critical damage threshold at time t, is the initial threshold, α is the acceleration factor determined by the microseismic energy growth slope, and t is the time factor. At the same time, the numerical model is divided into high damage gradient subdomains according to the damage acceleration area range, and the damage extension rate parameter of the subdomain is increased by a preset multiple to generate a local enhanced damage parameter signal, which is finally synchronously updated to the numerical model parameter signal to simulate the nonlinear acceleration characteristics of damage.

[0039] The stability assessment link performs dual-path processing based on the updated numerical model parameter signal: the physical path inputs the parameters into the strength reduction module, initializes the reduction coefficient to 1.0 and reduces the strength parameters with a fixed step size. After each reduction, a numerical simulation is run to monitor the rate of change of the global displacement field. When the displacement growth rate exceeds the divergence critical value, the current reduction coefficient is recorded as the safety factor signal; the data path extracts the maximum spatial gradient of the damage probability field, the volume proportion of the high-damage area, the energy release rate of the microseismic event, and the coefficient of variation of the principal strain direction of the strain field from the fusion state signal to construct an input vector, which is then sent to the survival analysis machine learning model trained with historical instability cases. , the proportional risk regression algorithm is used to output the failure probability signal of the future set time window; finally, the safety factor signal and the failure probability signal are input into the risk assessment matrix (the horizontal axis safety factor range is 0.8-2.0, and the vertical axis failure probability range is 0%-40%), and four risk areas are divided according to the preset boundaries and corresponding graded warning instructions are generated: when in area I, a green safety instruction is output; when entering area II, a yellow inspection instruction is output and the local high-frequency monitoring mode is triggered; when entering area III, an orange warning instruction is output and the emergency reinforcement plan generation module is activated; when entering area IV, a red evacuation instruction is output to link the mine dispatching system.

[0040] Furthermore, the online inversion algorithm uses the ensemble Kalman filter framework to realize dynamic parameter identification, and its process is as follows: define the cohesion parameter set, internal friction angle parameter set, damage threshold parameter set and damage extension rate parameter set of the numerical model to form a state vector signal; extract the three-dimensional space grid strain components from the fused state signal and construct the observation vector signal by sorting the node coordinates; set the parameter prior distribution based on the laboratory test value and generate the initial parameter set of a preset number of members through Latin hypercube sampling; input each set member into the numerical model and run the mechanical calculation in parallel to output the corresponding simulated strain field signal set; calculate the parameter set mean vector and covariance matrix, and simultaneously calculate the simulated strain field set. The combined observation variable covariance matrix and parameter-observation variable cross-covariance matrix are calculated; the difference between the actual observation vector signal and the simulated observation vector set is multiplied by the Kalman gain matrix to generate a parameter increment adjustment signal, where the Kalman gain matrix is ​​determined by the product of the cross-covariance matrix and the inverse of the observation covariance matrix; the root mean square error between the simulated strain field signal generated by the updated parameter set and the actual fusion state signal is iteratively calculated, and the calculation is terminated when the error change rate is less than the convergence threshold for three consecutive iterations, and the mean value of the current parameter set is output as the updated numerical model parameter signal; this signal is written into the numerical model database in real time through the application interface to replace the original material parameter group, completing the dynamic update of the constitutive relationship. During the extended implementation, a dynamic correction mechanism for damage evolution parameters is activated: the cumulative energy value per unit time of the three-dimensional spatial grid is calculated based on the microseismic event location signal in the fused state signal, and converted into a damage development degree indicator through the energy-damage mapping function; when it is detected that the growth rate of the damage indicator of the adjacent grid unit exceeds the preset acceleration threshold during the continuous monitoring period, a damage acceleration area coordinate signal is generated, and the standard linear attenuation function is reconstructed into an exponential attenuation function in the damage acceleration area. At the same time, according to the spatial range of the damage acceleration area, the high damage gradient subdomain is divided in the numerical model, and the damage propagation rate parameter of the subdomain material is increased by a preset multiple to generate a local enhanced damage parameter signal; finally, the updated damage critical threshold and damage propagation rate are synchronously integrated into the numerical model parameter signal, so that the model can accurately simulate the nonlinear acceleration effect of damage evolution.

[0041] Stability quantitative calculation is achieved through dual-path parallel processing: the physical path inputs the updated numerical model parameter signal into the strength reduction module, initializes the reduction coefficient to 1.0 and reduces the filling body strength parameter with a fixed step size (cohesion and internal friction angle are reduced synchronously), runs a numerical simulation after each reduction to monitor the rate of change of the global displacement field, and records the current reduction coefficient as a safety factor signal when the displacement growth rate exceeds the divergence critical value. If the critical value is not reached, the iteration continues until the maximum number of times; the data path extracts the maximum spatial gradient of the damage probability field, the volume proportion of the high-damage area, the energy release rate of the microseismic event per unit time, and the coefficient of variation of the principal strain direction of the strain field from the fusion state signal to construct an input vector, which is sent to the survival analysis machine learning model trained with historical instability cases. The model adopts The proportional hazards regression algorithm is used to establish the correlation between the characteristics and the instability time, and the conditional probability of instability within a set future time window is output as a failure probability signal. The two signals are fused through a risk assessment matrix: the safety factor signal is used as the horizontal axis (range 0.8-2.0) and the failure probability signal is used as the vertical axis (range 0%-40%). Four risk areas are divided: when the safety factor is ≥1.5 and the failure probability is ≤5%, it is judged as Area I (safe); when 1.2≤safety factor<1.5 or 5%<failure probability≤15%, it is judged as Area II (caution); when 1.0≤safety factor<1.2 or 15%<failure probability≤25%, it is judged as Area III (warning); when the safety factor is <1.0 or the failure probability is >25%, it is judged as Area IV (dangerous).

[0042] Specifically, the hierarchical warning instruction generation rules are implemented accordingly: when the comprehensive risk is in area I, a green safety instruction is output to maintain the normal operation of the system; when entering area II, a yellow inspection instruction is output and the local high-frequency monitoring mode is triggered (including increasing the distributed optical fiber sampling rate to three times the baseline value and activating the three-dimensional laser scanning device in the target area); when entering area III, an orange warning instruction is output and the emergency reinforcement plan generation module is activated. Based on the coordinate signal of the damage acceleration area and the current stress field distribution, the module outputs an anchor support density increase plan or a grouting reinforcement priority list; when entering area IV, a red evacuation instruction is output and the mine dispatching system is linked to automatically cut off the power supply to the dangerous area, start the escape channel indicator light sequence and the underground broadcast alarm system. In scenarios supported by edge computing architectures, the command execution process includes adaptive responses: In the green safety state, edge nodes switch microseismic sensors to a low-power sleep mode; in the yellow inspection state, they awaken dormant devices and increase the sampling rate of the first-class monitoring signal to twice the base frequency; in the orange alert state, underground robots equipped with portable resistivity imagers perform high-density scans of areas with accelerated damage; and in the red evacuation state, the personnel positioning system tracks the coordinates of those who have not evacuated in real time and sends the optimal escape route to mining smart helmets. All warning commands are dynamically mapped to the digital twin model via a 3D visualization engine: Green commands render the corresponding filling volume in a uniform blue; yellow commands superimpose a flashing yellow box in areas where the damage probability exceeds the threshold; orange commands further display red vortex markers in stress concentration areas; and red commands cover the model surface with a pulsating red light and indicate an evacuation countdown.

[0043] The mine filling performance detection and stability assessment system of the present invention includes an acquisition module, which uses a distributed optical fiber sensor array to collect strain and temperature spatial distribution signals in the filling in real time as a first type of monitoring signal, and simultaneously uses a microseismic sensor array to collect vibration waveform signals generated by internal rupture events in the filling as a second type of monitoring signal; a coordination module, which inputs the first type of monitoring signal and the second type of monitoring signal into a multi-source data fusion model to generate a fusion state signal that characterizes the three-dimensional damage degree and real-time deformation state of the filling; a verification module, which inputs the fusion state signal into a parameter dynamic identification module, calculates the strength parameters and damage evolution parameters of the filling through an online inversion algorithm, generates an updated numerical model parameter signal, and drives the mine filling numerical model to synchronously update its mechanical constitutive relationship; a preview module, which performs stability quantification calculation based on the updated numerical model parameter signal to generate a risk level signal, and the risk level signal is generated by fusing the safety factor output by the numerical simulation and the failure probability predicted by the machine learning model.

[0044] The implementation method of the edge computing and cloud collaborative architecture is as follows: an explosion-proof edge computing node is deployed on the side wall of the underground tunnel. The node integrates a multi-channel data acquisition card and a preprocessing unit, and connects the distributed optical fiber sensor array and the microseismic sensor array through an industrial bus; the edge node performs sliding window processing on the first type of monitoring signal collected in step S1, intercepts the original strain data of a time window every 10 seconds, extracts the frequency domain energy characteristics through wavelet transform and reduces the dimension to 20% of the original data volume to generate a compression strain characteristic signal; the second type of monitoring signal is preliminarily screened for events, and after filtering out microseismic events with energy less than 100 joules, the magnitude, location coordinates and duration of the remaining events are extracted to construct event feature vectors to generate compression microseismic characteristic signals; the two types of compression signals are processed by the mine The data is uploaded to the cloud server via a gigabit ring network. The multi-source data fusion model on the cloud performs reconstruction after receiving the signal: the compressive strain characteristic signal is restored to the full-resolution strain field using a generative adversarial network model, and the discrete microseismic event characteristic vector is reconstructed into a continuous energy density field through a spatial interpolation algorithm, thereby generating a high-fidelity fusion state signal; the parameter dynamic identification module and the stability quantification calculation module run in the cloud, and the generated risk level signal is encapsulated by the protocol and then transmitted to the edge node; the edge node has a built-in instruction conversion engine that parses the risk level signal into a device control instruction set, including distributed fiber optic demodulator sampling rate adjustment instructions, microseismic sensor wake-up instructions and 3D laser scanner trigger instructions, which are distributed to field equipment for execution via industrial real-time Ethernet.

[0045] Specifically, the adaptive monitoring strategy is as follows: when the risk level signal received by the edge node is a green safety instruction, the microseismic sensor array is controlled to switch to sleep mode (only the basic clock circuit is powered), and the distributed optical fiber sampling rate is reduced to 50% of the baseline value; when the risk level signal is upgraded to a yellow inspection instruction, the microseismic sensor is immediately awakened and a self-test pulse is sent to calibrate the timing, the optical fiber sampling rate is simultaneously increased to 200% of the baseline value, and the three-dimensional laser scanning device of the preset fan-shaped area on the surface of the filling body is activated. The device performs local point cloud acquisition at an interval of 5 seconds to generate surface displacement increments. If the risk level signal jumps to the orange warning command, the edge node starts the emergency scanning protocol: dispatching a wheeled inspection robot carrying a portable resistivity imager to the coordinate location of the damage acceleration area, deploying an electrode array at the target point to perform a cross scan, and generating a resistivity anomaly profile signal that is transmitted back to the cloud in real time; when the red evacuation command is triggered, the edge node switches to full-band monitoring mode: the optical fiber sampling rate is increased to the limit of 500Hz, the microseismic sensor starts the wide-band recording mode (0.1-5000Hz), and at the same time broadcasts the evacuation area geofence coordinates to the personnel positioning system. The 3D visualization engine implementation process is as follows: After receiving the fusion status signal, numerical model parameter signal, and risk level signal in the cloud, a 3D graphics interface is called to create a digital twin of the goaf. A damage probability field texture map is generated based on the gridded fusion status signal. The damage probability of 0-1.0 is converted into a blue-yellow-red gradient color spectrum through HSV color mapping, and then superimposed on the surface of the 3D grid model of the filling body to form a heat map signal. The principal strain direction field data is extracted from the fusion status signal, and a strain vector arrow cluster is drawn on the key monitoring section. The arrow length is proportional to the strain value and the direction is consistent with the principal strain direction, generating a deformation field direction indicator signal. The microseismic event location signal is converted into a spatiotemporal scattered point sequence. Each scattered point contains 3D coordinates, magnitude, and earthquake occurrence time attributes. The magnitude is mapped as the scattered point size, and the time is mapped as a transparency gradient, forming a spatial distribution signal of the microseismic event. The risk level signal drives the dynamic labeling of the scene: In the green safety state, the twin displays a translucent blue color. In the yellow inspection state, a flashing yellow wireframe is superimposed on the area where the damage probability exceeds the limit. In the orange alert state, a rotating red warning cone is added to mark the area of ​​accelerated damage. In the red evacuation state, a pulsed red light is activated to cover the model surface and display a countdown number. All signals are organized through a spatial index tree: an octree structure is used to manage the spatial grid of the goaf. Each node stores the mean damage probability, maximum strain value and microseismic event count of the corresponding area. The visualization engine dynamically schedules the LOD detail level based on the viewpoint position. When the view is focused, the full-resolution grid data is called to generate a high-precision rendering signal. In the perspective mode, it switches to aggregated statistical values ​​to generate a simplified rendering signal. The final output is a four-dimensional dynamic visualization stream containing heat map signals, deformation field direction signals, microseismic scatter point signals and risk marker signals, which is pushed to the terminal monitoring screen and mobile devices via the WebSocket protocol.

[0046] Furthermore, in scenarios where edge computing node resources are limited, a visualization preprocessing mechanism is activated: after receiving the risk level signal from the cloud, the edge node extracts the coordinate boundary of the damage acceleration area and calls a lightweight rendering engine to generate a local three-dimensional slice model; this model only contains grid data within 50 meters of the target area, and the texture map uses hierarchical compression technology (JPEG2000+regional interest coding), and vector arrows are simplified to direction icons; the generated simplified visualization signal is transmitted to the underground mobile terminal via a low-bandwidth channel (4G / LoRa) to ensure visual guidance of the evacuation route in an emergency; at the same time, the full model data on the cloud is continuously updated, and incremental synchronization is performed after the network is restored.

[0047] The mine filling performance detection and stability assessment method and system of the present invention generate a fusion status signal that characterizes three-dimensional damage by fusing multi-source heterogeneous signals of distributed optical fibers and microseismic arrays; utilize this signal to drive an online inversion algorithm to dynamically update numerical model parameters, so that the model approaches the real state in real time; generate graded warning instructions based on the updated model fusion safety factor and failure probability signal, forming a "perception-modeling-assessment" closed loop, overcoming the defects of traditional methods such as data isolation, static model, and delayed assessment.

[0048] Therefore, the mine filling body performance detection and stability evaluation method and system of the present invention can solve the problems of one-sided filling body status perception, model inaccuracy, and delayed warning.

[0049] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be encompassed by the present invention.

Claims

1. A method for detecting and evaluating the performance of mine fillings, characterized in that: include: S1: A distributed optical fiber sensor array is used to collect the spatial distribution signals of strain and temperature in the filling in real time as the first type of monitoring signal. At the same time, a microseismic sensor array is used to collect the vibration waveform signals generated by the internal rupture events of the filling as the second type of monitoring signal. S2: Inputting the first and second monitoring signals into a multi-source data fusion model to generate a fusion status signal representing the three-dimensional damage degree and real-time deformation status of the filling body, wherein the multi-source data fusion model performs feature extraction and spatial interpolation on heterogeneous temporal and spatial resolution data using a deep learning algorithm; S3: Inputting the fusion state signal into a parameter dynamic identification module, calculating the strength parameters and damage evolution parameters of the filling body through an online inversion algorithm, generating an updated numerical model parameter signal, and driving the mine filling body numerical model to synchronously update its mechanical constitutive relationship; S4: Based on the updated numerical model parameter signal, a stability quantification calculation is performed to generate a risk level signal. The risk level signal outputs a graded warning instruction by fusing the safety factor output by the numerical simulation and the failure probability predicted by the machine learning model.

2. The mine filling performance detection and stability assessment method according to claim 1, characterized in that: In step S1, a surface displacement field change signal is collected as a third type of monitoring signal by a three-dimensional laser scanning device arranged on the surface of the filling body, and a resistivity distribution signal is collected as a fourth type of monitoring signal by a geoelectric resistivity tomography device buried inside the filling body. In step S2, the third type of monitoring signal and the fourth type of monitoring signal are input into a multi-source data fusion model together with the first type of monitoring signal and the second type of monitoring signal to generate an enhanced fusion state signal that includes both the internal damage state and the surface deformation characteristics.

3. The mine filling performance detection and stability assessment method according to claim 1, characterized in that: The operations performed by the multi-source data fusion model in step S2 include: performing temperature compensation calculation on the first type of monitoring signal to eliminate the interference of ambient temperature on strain measurement, generating a temperature-corrected strain signal, performing waveform feature extraction on the second type of monitoring signal to identify the spatial location and energy level of the rupture event, generating a microseismic event location signal, mapping the temperature-corrected strain signal and the microseismic event location signal to a unified three-dimensional spatial grid through a spatiotemporal registration algorithm, using a convolutional neural network to perform cross-modal feature fusion on the grid node data, and outputting a fusion status signal that represents the probability of spatial distribution of damage.

4. The mine filling performance detection and stability assessment method according to claim 1, characterized in that: The online inversion algorithm described in step S3 adopts an ensemble Kalman filter framework to assimilate the fused state signal into the observation data, and iteratively adjusts the numerical model parameters to make the simulated output signal approach the observation signal. When the root mean square error between the simulated strain field signal and the strain component in the fused state signal is less than the set threshold, the strength parameter and damage evolution parameter of the current iteration step are output as the updated numerical model parameter signal.

5. The mine filling performance detection and stability assessment method according to claim 1, characterized in that: The damage evolution parameters include the critical threshold of the damage variable and the damage expansion rate. By real-time tracking the spatiotemporal aggregation characteristics of the microseismic event location signal in the fused state signal, the attenuation function of the critical threshold of the damage variable is dynamically modified, so that the updated numerical model can reflect the accelerated effect of filling damage accumulation.

6. The mine filling performance detection and stability assessment method according to claim 1, characterized in that: The stability quantification calculation described in step S4 includes parallel dual-path processing: the first path inputs the updated numerical model parameter signal into the strength reduction module, iteratively reduces the filling body strength parameter until the numerical model diverges, and outputs a safety factor signal based on limit equilibrium; the second path inputs the fusion state signal into the survival analysis model trained with historical cases, outputs a failure probability signal within a set time period in the future, and finally weightedly fuses the safety factor signal and the failure probability signal to generate a comprehensive risk level signal.

7. The mine filling performance detection and stability assessment method according to claim 1, characterized in that: The generation rules of the graded warning instructions are as follows: when the comprehensive risk level signal is in the first interval, a green safety instruction is output; when entering the second interval, a yellow inspection instruction is output and the local high-frequency monitoring mode is triggered; when entering the third interval, an orange warning instruction is output and the emergency reinforcement plan generation module is activated; when entering the fourth interval, a red evacuation instruction is output and the mine dispatching system is linked.

8. The mine filling performance detection and stability assessment method according to claim 1, characterized in that: It also includes step S5: deploying edge computing nodes underground, preprocessing the first and second monitoring signals collected in step S1, generating compressed feature signals and uploading them to the cloud, the multi-source data fusion model, parameter dynamic identification module and stability quantification calculation are executed in the cloud, and the generated risk level signal is transmitted to the edge computing node and converted into an early warning instruction.

9. The mine filling performance detection and stability assessment method according to claim 8, characterized in that: The edge computing node executes an adaptive monitoring strategy: when the received risk level signal is in the green safety instruction state, the sampling frequency of the first type of monitoring signal is reduced to the basic frequency; when the risk level signal is upgraded to the yellow inspection instruction state, the dormant microseismic sensor array is automatically awakened and a high-precision scanning instruction is triggered.

10. A mine filling performance detection and stability evaluation system, using the mine filling performance detection and stability evaluation method according to any one of claims 1 to 9, characterized in that: include: an acquisition module, which uses a distributed optical fiber sensor array to collect, in real time, spatial distribution signals of strain and temperature within the filling as a first type of monitoring signal, and simultaneously uses a microseismic sensor array to collect vibration waveform signals generated by rupture events within the filling as a second type of monitoring signal; a coordination module, wherein the coordination module inputs the first type of monitoring signals and the second type of monitoring signals into a multi-source data fusion model to generate a fusion state signal representing the three-dimensional damage degree and real-time deformation state of the filling body; A verification module, which inputs the fusion state signal into a parameter dynamic identification module, calculates the strength parameters and damage evolution parameters of the filling body through an online inversion algorithm, generates an updated numerical model parameter signal, and drives the mine filling body numerical model to synchronously update its mechanical constitutive relationship; A preview module, which performs stability quantification calculation based on the updated numerical model parameter signal to generate a risk level signal, wherein the risk level signal is generated by fusing the safety factor output by the numerical simulation with the failure probability predicted by the machine learning model.

Images