A kind of open-pit slope sliding trend early warning system based on active micro-vibration excitation

Abstract

The present application relates to the technical field of open-pit mine slope sliding trend early warning, and specifically discloses an open-pit mine slope sliding trend early warning system based on active micro-vibration excitation, which comprises the following components: an excitation unit composed of adjustable frequency micro-vibration exciters, which periodically apply low-amplitude controllable mechanical disturbance signals to the slope rock mass, excite the response of the rock-soil medium structure and improve the early warning sensitivity; a sensing unit comprising a three-axis acceleration sensor, a displacement meter, a GNSS sensor and a ground sound wave sensor, which records the response changes in the propagation process of the vibration disturbance signal and obtains data; an analysis unit comprising an edge intelligent terminal, a sliding trend identification model built in the edge intelligent terminal and a lightweight neural network, wherein the sliding trend identification model performs sensing processing on the data and outputs a sliding trend label; and a communication control unit for ensuring the stable data flow of the above units and the collaborative operation of the modules.

Description

Technical Field

[0001] This application relates to the field of early warning technology for slope slippage trends in open-pit mines, and specifically discloses an early warning system for slope slippage trends in open-pit mines based on active micro-vibration excitation. Background Technology

[0002] Open-pit mine slope stability monitoring and early warning technology, as an important component of the mine safety production system, has undergone decades of theoretical evolution and engineering practice, forming a technical framework centered on geological surveys, engineering monitoring, numerical analysis, and risk early warning. With the continuous increase in the mining depth and slope angle of open-pit mines, slope stability issues are becoming increasingly prominent. The resulting landslides, collapses, and other geological disasters not only threaten personnel safety but also seriously disrupt production continuity. Slope disasters are characterized by their suddenness, concealed development process, and short early warning response time, necessitating the establishment of a monitoring and early warning system with higher sensitivity, real-time performance, and reliability.

[0003] Traditional monitoring methods primarily rely on point-based sensors such as total stations, GNSS, crack gauges, and inclinometers to periodically monitor surface or localized slope displacement. While capable of identifying some gradually changing slip risks, these methods suffer from response lag, monitoring blind spots, and data fragmentation when facing sudden, deep, or structural instability. In recent years, non-contact monitoring technologies such as laser scanning, ground-based synthetic aperture radar (GB-SAR), and UAV surveying have been applied, improving spatial coverage and data acquisition efficiency. However, these technologies are limited by climate conditions, equipment costs, and the complexity of data interpretation, making it difficult to completely replace traditional sensing methods.

[0004] In terms of data processing, slope stability analysis methods are evolving from empirical models and limit equilibrium methods to numerical simulation and intelligent analysis. Some studies have incorporated artificial intelligence technologies, such as support vector machines, neural networks, and decision trees, for deformation trend prediction and landslide severity assessment. However, due to issues such as unstable input data quality, missing data labels, and insufficient model generalization ability, their engineering applications are still in the exploratory stage.

[0005] In terms of monitoring data processing, slope stability analysis methods are evolving from empirical models and limit equilibrium methods to numerical simulation and intelligent analysis. Some studies have introduced artificial intelligence technologies, such as support vector machines, neural networks, and decision trees, for deformation trend prediction and landslide severity assessment. However, due to problems such as unstable input data quality, missing data labels, and insufficient model generalization ability, its engineering application is still in the exploratory stage.

[0006] Most existing slope early warning systems rely on passive response signals, that is, they judge the rock mass response behavior triggered by natural factors (such as rainfall, mining, earthquakes). This passive mode has shortcomings such as strong dependence on the time of deformation occurrence, insufficient prediction lead time, and difficulty in matching the monitoring cycle with the sliding evolution process, making it difficult to identify deep, slowly changing sliding risks in advance.

[0007] To address the aforementioned issues, the concept of "active excitation-response identification" has gradually emerged in the field of geotechnical engineering and structural health monitoring in recent years. This method applies low-intensity disturbances to the target structure through external excitation, and extracts changes in structural stiffness or mechanical properties by combining the frequency, wave velocity, and attenuation characteristics of the response signal, thereby achieving condition assessment and early warning. Although this method has been preliminarily validated in structural engineering projects such as bridges and dams, its application in the field of mine slope monitoring is still in its early stages. Related theoretical research and technical approaches are not yet mature, especially lacking system integration and dynamic early warning solutions for open-pit mining conditions.

[0008] Currently, the existing technology has the following problems:

[0009] Passive monitoring relies on natural triggers, resulting in delayed responses and insufficient predictive lead time. Existing slope stability monitoring technologies generally employ a passive monitoring mode, using sensors such as GNSS, crack gauges, and inclinometers to capture response signals generated by slopes under naturally induced conditions such as rainfall, mining, and earthquakes, thereby identifying potential instability trends. However, this type of technology is highly dependent on natural triggers and cannot actively intervene in the slope state change process, leading to insufficient data timeliness, susceptibility to environmental noise interference, and difficulty in identifying early, weak signs before significant slippage has developed, thus limiting the foresight and accuracy of early warnings.

[0010] The limited level of multi-source monitoring data fusion fails to reflect the overall stiffness evolution of slopes. Although some monitoring systems have attempted to incorporate non-contact technologies such as laser scanning, ground-based radar, and UAV mapping, and combine them with point-based sensor data for information fusion, most systems only focus on surface or local displacement information of the slope, lacking a parameter system capable of characterizing the overall structural mechanical performance. In the process of multi-source data fusion analysis, a standardized stiffness / softening index system has not yet been established, and most rely solely on deformation itself for risk assessment, failing to quantify the internal evolution process of the slope's mechanical response. Especially when facing slowly changing, deep-seated slip risks, existing technologies struggle to provide quantitative trends in stability evolution, limiting their applicability in early slope disaster identification.

[0011] The lack of controllable excitation methods leads to unstable signal excitation and discontinuous evaluation mechanisms: Currently, microseismic signals used for slope monitoring in mining engineering are mostly caused by natural micro-fractures, mining disturbances, or blasting operations. These signals are uncontrollable and non-periodic, resulting in uneven distribution of observation data over time and a lack of continuity and repeatability in evaluation indicators. Furthermore, environmental noise and high-frequency interference can easily mask useful microseismic information, affecting the analysis results. In addition, existing schemes generally do not incorporate the "active excitation-response analysis" technique widely used in structural health monitoring, making it difficult to effectively assess deep mechanical changes such as rock mass stiffness degradation and structural damage.

[0012] Limited methods exist for modeling slip trends, and intelligent early warning models suffer from weak generalization ability: Some studies have attempted to use artificial intelligence methods such as neural networks and support vector machines to analyze trends and classify the severity of slope deformation data. However, due to insufficient training samples, limited feature dimensions, and difficulty in matching model complexity to the dynamic changes in mining sites, the results are prone to overfitting, false alarms, or missed alarms. Currently, no intelligent modeling system integrating active excitation signal characteristics and multi-source sensor data has been established, making it impossible to fully utilize the deep evolutionary characteristics in time-series response data for dynamic modeling and risk quantification of the slip process.

[0013] In summary, existing technologies have made some progress in the monitoring and early warning of open-pit mine slope stability, but problems still exist, such as limited monitoring methods, insufficient data processing capabilities, and delayed early warning response. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation proposed in this invention aims to improve the scientific rigor and effectiveness of slope disaster prevention and control by introducing an active excitation mechanism, combining multi-source sensor data and intelligent analysis models to achieve real-time monitoring of slope stability and early warning of slippage trends.

[0014] In view of this, the present invention provides an early warning system for the slippage trend of open-pit mine slopes based on active micro-vibration excitation, so as to solve the above problems. Summary of the Invention

[0015] This invention aims to address the problems existing in current open-pit mine slope stability monitoring and early warning methods, such as response lag, insufficient early warning lead time, difficulty in quantifying mechanical evolution, and weak intelligent analysis capabilities. It constructs a slope slip early warning system with active excitation capability, continuous response monitoring, and intelligent trend recognition, so as to achieve early identification and graded response to the slope structure stability change process.

[0016] To achieve the above objectives, the present invention provides the following basic technical solution:

[0017] An early warning system for landslide trends on open-pit mine slopes based on active micro-vibration excitation includes the following components:

[0018] Excitation unit: Composed of several adjustable frequency micro-vibration exciters, which periodically apply low-amplitude controllable mechanical disturbance signals to the slope rock mass. , To stimulate the structural response of soil and rock media and improve early warning sensitivity;

[0019] Sensing unit: includes a triaxial accelerometer ( ), displacement gauge ( GNSS sensor ( ) and ground acoustic wave sensor ( Record vibration disturbance signals Response changes during propagation and data acquisition ;

[0020] Analysis Unit: Includes an edge intelligent terminal, a slip trend recognition model built into the edge intelligent terminal, and a lightweight neural network. The slip trend recognition model analyzes the data. Perform sensor processing and analyze the data at each time step. Classify and output slip trend labels;

[0021] Interactive unit: including command center platform, field mobile terminal and remote data management system, which receives and parses the early warning information output by slip trend recognition model and then executes response;

[0022] Communication control unit: Used to ensure stable data flow of the above units and coordinated operation of each module.

[0023] Furthermore, the tunable frequency micro-vibration exciter, by anchoring itself to the slope surface, the toe of the slope, and the stabilizing support, disturbs the signal. The excitation controller generates the data in the form of a function, as follows:

[0024] ;

[0025] in, : Initial amplitude of the excitation; Excitation frequency; Initial phase; : Time variable.

[0026] Furthermore, the triaxial accelerometer ( ), displacement gauge ( GNSS sensor ( ) and ground acoustic wave sensor ( Generate the original data vector , ,in, Triaxial acceleration; Displacement; GNSS plane offset; Ground acoustic signal response, all sensors use the same sampling frequency. The timestamp is synchronously distributed by the communication control unit.

[0027] Furthermore, the lightweight neural network is ,definition , The fused feature vector, generated by the data fusion module, represents the slope slippage trend classification result: 0 = safe, 1 = weak anomaly, and 2 = strong anomaly.

[0028] Furthermore, the main control module that controls the adjustable frequency micro-vibration exciter issues a unified timestamp command. A frequency-adjustable micro-vibration exciter initiates micro-vibration disturbances, and a triaxial accelerometer ( ), displacement gauge ( GNSS sensor ( ) and ground acoustic wave sensor ( )exist to Data is collected synchronously within the time window, including: The edge intelligent terminal represents the response duration window. It performs preliminary filtering, threshold removal, and event marking on the collected signals, numbering the valid responses under each stimulus as follows: ,in The stimulus-response cycle sequence numbers are as follows:

[0029] ;

[0030] in, This represents the triaxial values ​​acquired by each acceleration channel; Represents the GNSS displacement response vector; This response data represents the change in angle. The data is then uploaded to an edge intelligent terminal as input features for the slip trend recognition model.

[0031] Furthermore, after the response data enters the edge intelligent terminal, it is necessary to perform synchronous fusion and unified formatting of multi-source sensor signals to construct a high-dimensional feature vector for slip trend recognition. The specific steps are as follows:

[0032] S01: The edge intelligent terminal embeds a data fusion and feature extraction module, which is triggered by an incentive time. Based on this, the alignment, cleaning, and feature extraction of different physical quantity data within a time window are achieved;

[0033] S02: Perform bandpass filtering and normalization preprocessing on the input data from each sensor: for acceleration response Eliminate background noise and low-frequency drift; for displacement data Apply the moving average to remove platform drift;

[0034] S03: All preprocessed signals are sampled at a uniform frequency. Interpolation resampling is performed below, and in the window Inner alignment to form a uniform response matrix:

[0035] ;

[0036] in: For time series points, Indicates the first The multidimensional response fusion matrix corresponding to the secondary stimulus;

[0037] S04: A composite feature set is constructed using three types of indices: time domain, frequency domain, and nonlinear dynamics. Each excitation response data is mapped to a feature vector of fixed length.

[0038] ;

[0039] in Indicates the first Each feature extraction function in the fusion matrix The result of the operation is the total feature dimension.

[0040] Furthermore, the slip trend recognition model uses feature vectors As input, output the slip trend level of the slope under its current condition. , respectively representing "stable state", "gradually changing state" and "critical glide state";

[0041] The method for constructing a slip trend recognition model is as follows:

[0042] A01: A supplementary modeling approach using a random forest model as the primary identification framework, supplemented by a lightweight neural network for predicting slippage trend evolution. The random forest model is composed of... It consists of decision trees, each trained using random feature subsets and sample subsets, and the final prediction result is obtained by the voting integration of all trees;

[0043] A02: Obtaining a slip trend recognition model:

[0044] ;

[0045] in, Indicates the first The decision function for a tree, where mde{} is the mode function;

[0046] Each decision tree is divided into nodes based on the Gini index or entropy increase criterion, thereby constructing a layer-by-layer judgment path for the slip state.

[0047] Furthermore, the slip trend recognition model uses feature vectors As input, output the slip trend level of the slope under its current condition. , respectively representing "stable state", "gradually changing state" and "critical glide state";

[0048] The method for constructing a slip trend recognition model is as follows:

[0049] A01: A supplementary modeling approach using a random forest model as the primary identification framework, supplemented by a lightweight neural network for predicting slippage trend evolution. The random forest model is composed of... It consists of decision trees, each trained using random feature subsets and sample subsets, and the final prediction result is obtained by the voting integration of all trees;

[0050] A02: Obtaining a slip trend recognition model:

[0051] ;

[0052] in, Indicates the first The decision function for a tree, where mde{} is the mode function;

[0053] Each decision tree is divided into nodes based on the Gini index or entropy increase criterion, thereby constructing a layer-by-layer judgment path for the slip state.

[0054] Furthermore, the early warning mechanism is as follows:

[0055] The early warning mechanism uses the prediction results and state score indicators output by the slip state recognition model as its core trigger. The mechanism includes tiered response and multi-channel dissemination, where tiered response involves setting a state score. The lower limit of the threshold and Upper limit, according to The lower limit of the threshold and The upper limit achieves the corresponding response. When the system continuously collects m excitation responses, if the m consecutive responses are judged to the same high-level state, the early warning locking mechanism is activated, the early warning command is synchronized to the edge intelligent terminal and the interaction unit, and a digital early warning file is generated. The digital early warning file includes the excitation number, response characteristics, judgment level and slope code.

[0056] Furthermore, the multi-channel release includes a local audible and visual alarm device, a control room display module, and a mobile client. The local audible and visual alarm device issues an audible and visual alarm in the slope operation area, the control room display module displays a real-time pop-up alarm on the GIS interface, and sends SMS or push notifications. The mobile client displays the warning simultaneously on a mobile app and web platform, enabling remote control.

[0057] The principle and effect of this solution are as follows:

[0058] 1. Compared with existing technologies, this system consists of a micro-vibration excitation unit, a response acquisition subsystem, a data fusion and feature extraction module, a slip trend identification model, an edge computing terminal, and a multi-terminal information dissemination module. The system deploys low-frequency micro-vibration exciters at key structural locations on the slope to periodically apply directional disturbance signals, while simultaneously acquiring acceleration, displacement, and GNSS response data of the slope structure to the excitation. Slip trend identification is then performed based on a random forest or neural network model. The early warning model runs locally on the edge computing device, enabling real-time stability scoring and tiered early warning information dissemination. Experimental results show that this system can effectively identify weak slip precursor signals in typical high slope environments, possessing advantages such as flexible deployment, strong adaptability, and timely early warning. This system is suitable for intelligent safety early warning needs in complex conditions such as steep slopes and mining areas with goaf.

[0059] 2. Compared with existing technologies, traditional slope monitoring methods mainly rely on passive response signal acquisition. Limited by natural triggering factors and the sensing range of equipment, they struggle to capture subtle deformation signs in a timely manner before significant slippage develops. The acquisition of early warning signals is random and delayed, causing slope risk identification to lose its first-mover advantage in engineering decision-making. This invention introduces a periodically controllable low-amplitude micro-vibration excitation device to apply disturbance signals within a safe range to the slope. By deploying multi-point response acquisition units in key structural areas, a "stimulus-response" analysis chain is constructed. This allows for the continuous and controllable acquisition of slope structure response information to disturbances, thereby proactively revealing changes in structural stiffness, local softening trends, and potential slippage signs.

[0060] 3. Compared with existing technologies, the system of this invention has high engineering feasibility, low energy consumption and flexible deployment of the excitation device, strong compatibility between the response acquisition equipment and existing mine monitoring networks, and the data processing and analysis module can be embedded in an edge computing platform, making it suitable for the continuous monitoring and dynamic risk management needs of open-pit mines. Through the deployment of this system, the response speed and prediction accuracy of slope disaster early warning can be effectively improved, providing a more forward-looking safety guarantee for mine operations. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This paper presents a schematic diagram of the structure of an open-pit mine slope slip trend early warning system based on active micro-vibration excitation, as proposed in an embodiment of this application.

[0063] Figure 2 This paper illustrates an early warning information flow diagram in an open-pit mine slope slip trend early warning system based on active micro-vibration excitation, as proposed in an embodiment of this application.

[0064] Figure 3 The diagram shows a multi-terminal deployment structure of an open-pit mine slope slip trend early warning system based on active micro-vibration excitation proposed in an embodiment of this application. Detailed Implementation

[0065] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0066] Implementation, for example Figures 1-3 As shown:

[0067] An early warning system for landslide trends on open-pit mine slopes based on active micro-vibration excitation includes the following components:

[0068] Excitation unit: Composed of several adjustable frequency micro-vibration exciters, which periodically apply low-amplitude controllable mechanical disturbance signals to the slope rock mass. , To stimulate the structural response of soil and rock media and improve early warning sensitivity;

[0069] Specifically: The adjustable frequency micro-vibration exciter is anchored to the slope surface, slope toe, and stabilizing support to disturb the signal. The excitation controller generates the data in the form of a function, as follows:

[0070] ;

[0071] in, : Initial amplitude of the excitation; Excitation frequency; Initial phase; Time variable;

[0072] Sensing unit: includes a triaxial accelerometer ( ), displacement gauge ( GNSS sensor ( ) and ground acoustic wave sensor ( Record vibration disturbance signals Response changes during propagation and data acquisition ;

[0073] Specifically: the triaxial accelerometer ( ), displacement gauge ( GNSS sensor ( ) and ground acoustic wave sensor ( Generate the original data vector , ,in, Triaxial acceleration; Displacement; GNSS plane offset; Ground acoustic signal response, all sensors use the same sampling frequency. The timestamp is synchronously distributed by the communication control unit.

[0074] Analysis Unit: Includes an edge intelligent terminal, a slip trend recognition model built into the edge intelligent terminal, and a lightweight neural network. The slip trend recognition model analyzes the data. Perform sensor processing and analyze the data at each time step. Classify and output slip trend labels;

[0075] Specifically: the lightweight neural network is ,definition , The fused feature vector, generated by the data fusion module, represents the slope slippage trend classification result: 0 = safe, 1 = weak anomaly, and 2 = strong anomaly.

[0076] Interactive unit: including command center platform, field mobile terminal and remote data management system, which receives and parses the early warning information output by slip trend recognition model and then executes response;

[0077] The interaction layer consists of a command center platform, field mobile terminals, and a remote data management system, forming a complete closed loop of "early warning generation—push—response." Through the control unit, each terminal receives and parses the early warning information. It also performs: information push, emergency notification linkage, and user feedback and correction.

[0078] Communication Control Unit: Used to ensure the stability of data flow of the above units and the coordinated operation of each module. In order to ensure the stability of data flow and the coordinated operation of each module, the system sets up a communication control unit, which is responsible for functions such as clock synchronization, data encoding, edge buffering, remote transmission and network fault tolerance. The communication protocol adopts MQTT as the main protocol, supplemented by HTTPS to achieve dual-channel encryption.

[0079] Regarding the deployment of related components:

[0080] In a typical open-pit mine slope scenario, tunable frequency micro-vibration exciters are deployed in the lower part of the slope, making close contact with the structure; various sensors are deployed along the slope in an equidistant linear manner, covering different areas of the rock; edge intelligent terminals are deployed adjacent to the sensors and are powered by solar energy; the command center is located in the dispatch room, and mobile terminals can be installed and run on the mobile phones, tablets and other devices of construction personnel to realize a full-area, real-time slippage early warning and perception network;

[0081] The main control module that controls the adjustable frequency micro-vibration exciter issues a unified timestamp command. A frequency-adjustable micro-vibration exciter initiates micro-vibration disturbances, and a triaxial accelerometer ( ), displacement gauge ( GNSS sensor ( ) and ground acoustic wave sensor ( )exist to Data is collected synchronously within the time window, including: The edge intelligent terminal represents the response duration window. It performs preliminary filtering, threshold removal, and event marking on the collected signals, numbering the valid responses under each stimulus as follows: ,in The stimulus-response cycle sequence numbers are as follows:

[0082] ;

[0083] in, This represents the triaxial values ​​acquired by each acceleration channel; Represents the GNSS displacement response vector; This response data represents the angular change obtained via a displacement gauge. The data is then uploaded to an edge intelligent terminal as input features for the slip trend recognition model.

[0084] Specifically as follows:

[0085] The functions of each type of sensor are divided as follows: Accelerometer Deployed in boreholes on the surface and inside the slope, the sensor captures local acceleration changes caused by micro-vibration excitation. It supports sampling frequencies up to 500 Hz and a minimum resolution of 0.001 m / s², primarily recording vertical and tangential disturbance amplitudes to construct the dynamic response spectrum of the slope medium.

[0086] GNSS displacement sensor Installed on slope surfaces and crack-sensitive areas, it records the cumulative displacement of minute movements. The acquisition frequency is typically set to 1Hz, and combined with real-time differential (RTK) technology, it can achieve an accuracy of ±1cm, outputting a two-dimensional displacement vector. .

[0087] Tilt sensor : Embedded deep within the rock mass or in anchor holes at the toe of the slope, it collects real-time data on changes in the slope's dip angle. It can determine micro-slippage and creep trends with a measurement accuracy better than ±0.05°.

[0088] In terms of deployment strategy, sensor density and placement points are adjusted according to the slope's hazard level. In high-risk areas, a complete set of sensor units is deployed every 10 meters, while in the next highest-risk areas, a sparser deployment is used, and the placement scheme is optimized by fusing historical deformation data. To ensure long-term system reliability, the acquisition units are equipped with functions such as automatic time synchronization, low-voltage alarm, breakpoint resume, and data compression.

[0089] After the response data enters the edge intelligent terminal, it is necessary to perform synchronous fusion and unified formatting of multi-source sensor signals to construct a high-dimensional feature vector for slip trend recognition. The specific steps are as follows:

[0090] S01: The edge intelligent terminal embeds a data fusion and feature extraction module, which is triggered by an incentive time. Based on this, the alignment, cleaning, and feature extraction of different physical quantity data within a time window are achieved;

[0091] S02: Perform bandpass filtering and normalization preprocessing on the input data from each sensor: for acceleration response Eliminate background noise and low-frequency drift; for displacement data Apply the moving average to remove platform drift;

[0092] S03: All preprocessed signals are sampled at a uniform frequency. Interpolation resampling is performed below, and in the window Inner alignment to form a uniform response matrix:

[0093] ;

[0094] in: For time series points, Indicates the first The multidimensional response fusion matrix corresponding to the secondary stimulus;

[0095] S04: A composite feature set is constructed using three types of indices: time domain, frequency domain, and nonlinear dynamics. Each excitation response data is mapped to a feature vector of fixed length.

[0096] ;

[0097] in Indicates the first Each feature extraction function in the fusion matrix The result of the operation is the total feature dimension.

[0098] Regarding time-domain features: such as mean, variance, skewness, kurtosis, peak factor, etc.; frequency-domain features: extracting the dominant frequency, frequency band energy distribution, spectral centroid, etc. through Fast Fourier Transform (FFT); nonlinear features: including sample entropy, wavelet entropy, Hurst exponent, etc., used to characterize signal complexity and long-term dependence of system state.

[0099] The slip trend recognition model uses feature vectors As input, output the slip trend level of the slope under its current condition. , respectively representing "stable state", "gradually changing state" and "critical glide state";

[0100] The method for constructing a slip trend recognition model is as follows:

[0101] A01: A supplementary modeling approach using a random forest model as the primary identification framework, supplemented by a lightweight neural network for predicting slippage trend evolution. The random forest model is composed of... It consists of decision trees, each trained using random feature subsets and sample subsets, and the final prediction result is obtained by the voting integration of all trees;

[0102] A02: Obtaining a slip trend recognition model:

[0103] ;

[0104] in, Indicates the first The decision function for a tree, where mde{} is the mode function;

[0105] Each decision tree is divided into nodes based on the Gini index or entropy increase criterion, thereby constructing a layer-by-layer judgment path for the slip state.

[0106] Specifically:

[0107] Input feature vector of the slip trend recognition model and output slip trend level To train the model and build a training sample set: The total capacity of the sample library is Once the slip trend recognition model is deployed, it generates real-time predictive output after each round of excitation. A stability determination mechanism for the slip state was designed. When the slip state score exceeds the set threshold If so, the system will automatically trigger the early warning mechanism;

[0108] The early warning mechanism is as follows: The early warning mechanism uses the prediction results and state score indicators output by the slip state recognition model as its core trigger. The early warning mechanism includes tiered response and multi-channel dissemination, namely… Figure 2The multi-channel early warning release module includes a tiered response system with status scoring. The lower limit of the threshold and Upper limit, according to The lower limit of the threshold and The upper limit achieves the corresponding response. When the system continuously collects m excitation responses, if the m consecutive responses are judged to the same high-level state, the early warning locking mechanism is activated, the early warning command is synchronized to the edge intelligent terminal and the interaction unit, and a digital early warning file is generated. The digital early warning file includes the excitation number, response characteristics, judgment level and slope code.

[0109] The system classifies the early warning levels into three levels based on the urgency of the slope slippage trend, as shown in Table 1 below:

[0110]

[0111] Table 1

[0112] in, and These are the lower and upper limits of the state scoring threshold, respectively, typically 0.8 and 1.5, and can be adaptively adjusted according to the actual slope sensitivity.

[0113] Specifically: such as Figure 2 As shown, the warning levels are divided into three levels: State 1, State 2, and State 3. State 1, State 2, and State 3 correspond to entering data archiving, triggering the Level 1 warning process, and triggering the Level 3 warning process, respectively. After the system triggers a high-level warning, it calculates the risk level and then enters the information integration module, proceeding to the following chain response process:

[0114] Status confirmation: The system continuously collects data. The stimulus response, if continuous If the same high-level status is determined, an early warning and locking mechanism will be activated to avoid false alarms.

[0115] Task distribution: The warning instructions are synchronized to the edge intelligent terminal and the control center host, and a digital warning file is generated. The digital warning file contains information such as the excitation number, response characteristics, judgment level, and slope code.

[0116] Multi-channel publishing refers to the multi-channel publishing module: Local audible and visual alarm device: issues audible and visual alarms in the slope operation area;

[0117] Control room display module: Real-time pop-up alarms on the GIS interface, and sending SMS or push notifications;

[0118] Mobile client: Alerts are displayed simultaneously on mobile apps and web platforms, enabling remote control.

[0119] Triggering of coordinated measures: For Level 1 warnings, the system can activate emergency plans, such as dispatching vehicles for evacuation and temporarily sealing off dangerous areas on slopes.

[0120] Retrospective archiving: Each warning event automatically generates a complete log, including time, status sequence, operation records, etc., for subsequent post-event analysis and model verification. Finally, the response closed-loop confirmation module is activated to close the response loop.

[0121] To ensure the timeliness and reliability of early warning information, the system employs a multi-level caching and link redundancy mechanism:

[0122] The local early warning control terminal possesses autonomous judgment and release capabilities, and can independently complete information distribution even if it loses connection with the central control center. Data transmission adopts a dual-channel mechanism of HTTPS and MQTT to ensure multi-point encrypted communication between the edge, center, and terminal. The early warning release log has digital signatures and anti-tampering mechanisms, supporting system supervision and accountability. Through the above mechanisms, this invention can realize a real-time, accurate, and traceable early warning response process for open-pit mine slopes, ensuring accurate information transmission and effective command execution in critical situations, significantly improving the safety level of high-slope operation in open-pit mines.

[0123] To achieve low-latency response and on-site autonomy of the system, this application introduces an edge intelligent terminal into the slope monitoring system. Its core functions are: on-site excitation control, response data fusion processing, slippage trend identification, and preliminary early warning issuance.

[0124] The edge intelligent terminal is installed in a safe location near the slope operation site. It adopts an industrial-grade low-power embedded host, equipped with an AI computing module and a communication gateway, and has the following core components:

[0125] Signal-driven controller: controls the micro-vibration excitation device to emit a specified waveform excitation signal;

[0126] Multi-source data acquisition interface: connects to devices such as accelerometers, GNSS, displacement gauges, and pore water pressure sensors;

[0127] Local early warning management module: Triggers local audible and visual alarms and uploads early warning information based on model judgment results and early warning level strategies;

[0128] Communication management module: Enables stable communication with the central server and mobile devices via MQTT+HTTP protocol.

[0129] The edge intelligent terminal supports both ARM and x86 architecture deployment modes and runs a lightweight Linux system, ensuring system portability and adaptability to the field environment.

[0130] like Figure 1As shown, the excitation control module A1 is activated, i.e., the excitation unit is activated; the response acquisition module A2 is activated, i.e., the perception unit is activated; the data fusion and feature extraction module A3 and the slip trend recognition module A4 are activated, i.e., the slip trend recognition model in the analysis unit analyzes the data. Perform sensor processing and analyze the data at each time step. The system categorizes data, outputs sliding trend labels, and the early warning management module A5 archives and generates logs for the early warning mechanism. For example... Figure 3 As shown: Information push notifications via mobile app, web platform, buzzer, and dispatch center are supported by technology via WiFi, 4G, or MQTT; Figure 3 In this module, the micro-vibration excitation unit is the micro-vibration excitation control module, and the response data acquisition module is the sensing unit.

[0131] Example 1: Application of Active Stimulation-Response Monitoring for Typical Mine Slopes

[0132] In the northern high slope area of ​​a metal mine open-pit mine, there is a steep rock slope with a height of approximately 65 meters and a dip angle of approximately 48°. The geological structure is mainly composed of layered gneiss with well-developed fissures, posing a certain risk of slippage. To demonstrate the application of the system of this invention, an active micro-vibration excitation and multi-source response acquisition system was deployed on this slope, and continuous monitoring tests were conducted.

[0133] The excitation device was deployed at the slope shoulder and near two high-risk joint zones, consisting of three sets of electrically driven vibration units (numbered E1, E2, and E3). The excitation frequency was adjustable from 20 to 80 Hz, and the excitation power was controlled within the range of 50 to 150 N. Micro-vibration pulses were applied cyclically through the controller, completing a full excitation cycle every 10 minutes. The excitation signal propagated downwards after coupling through the rock mass, forming a quantifiable elastic response field.

[0134] Downstream of the excitation path, deploy response sensor networks, with a total of:

[0135] Three sets of triaxial accelerometers (numbered S1 to S3), with a sampling frequency of 1000Hz;

[0136] Two sets of GNSS high-precision displacement stations (numbered G1 and G2) have a measurement accuracy better than 2 mm.

[0137] One set of fiber optic displacement gauge array (numbered D1) covers a sliding surface length of approximately 45 meters.

[0138] All sensors aggregate data through an edge computing terminal, which performs time synchronization, filtering, spectrum conversion, and extracts key features such as acceleration amplitude envelope, frequency band energy distribution, and coupling delay changes.

[0139] The extracted feature vectors are input into the slip trend recognition model established in this invention. The trained random forest algorithm (tree depth = 5, number of trees = 100) outputs a risk level score for the current stable state. The system sets a threshold T1 = 0.72. When the output score... At that time, the slope was determined to have entered a critical state of potential slippage.

[0140] During a 72-hour continuous field test, the system identified three moderate-intensity slope stress concentration responses, with model output scores of 0.63, 0.59, and 0.78, respectively. One of these responses exceeded a preset threshold, triggering an audible and visual warning and sending an alarm message to the management terminal. Subsequent investigation revealed that this period coincided with a period of heavy rainfall followed by increased rock mass moisture content and reduced frictional resistance on the sliding surface, demonstrating a high degree of consistency between the system's identification results and the actual hazard trend.

[0141] This first embodiment verifies the deployability, stability, and early warning accuracy of the system of the present invention in a real mining environment, and has good prospects for engineering application and promotion.

[0142] Example 2: Adaptive training process of model under critical conditions of typical slope:

[0143] To improve the accuracy and adaptability of slip trend identification, this invention establishes an adaptive training mechanism based on microvibration response data collected on-site at the slope, in order to construct a slip trend identification model applicable to different geological structures and excitation paths. This process constructs classification labels based on known state samples and employs supervised learning methods to optimize model parameters.

[0144] Taking a typical open-pit metal mine slope in eastern China as an example, the slope is approximately 72 meters high, with moderately weathered gneiss as the main lithology and well-developed fissures. Three sets of excitation devices (denoted as E1–E3) were deployed, along with accelerometers S1–S4 and displacement gauges G1 and G2, forming a response acquisition subsystem. The excitation frequency was set to 40Hz, and the excitation period was 10 minutes.

[0145] The collected data was processed synchronously, and the following main feature vectors were extracted:

[0146] Acceleration signal main frequency band energy (unit: dB)

[0147] Propagation delay from stimulus to response (unit: ms)

[0148] Mean value of sensor node response amplitude envelope (unit: m / s²)

[0149] GNSS displacement rate of change (unit: mm / h)

[0150] Constructing a training sample set Each sample Corresponding to manually labeled stability tags , where 0 represents a stable state and 1 represents entering the slip critical region.

[0151] The model was trained using a Random Forest (RF) classifier, with the forest size set. Decision trees, maximum depth The model performance was evaluated using 10-fold cross-validation, and the following metrics were obtained on the validation set:

[0152] Accuracy: 92.6%;

[0153] Precision: 88.4%;

[0154] Recall rate: 90.1%;

[0155] F1 score: 89.2%;

[0156] After training, the model is deployed to edge devices to process newly acquired sample streams in real time and output risk scores for each time period. ,when (The threshold is set to 0.75) This triggers an early warning. In subsequent operation, the system can dynamically identify the trend evolution reflected by changes in structural response, improving the timeliness and practicality of slip prediction.

[0157] In addition, to ensure that the model remains effective despite long-term changes in the site, the system integrates an online fine-tuning mechanism. When the collected samples meet the update conditions, the retraining process is triggered to update the model parameters and achieve long-term adaptive operation.

[0158] This second embodiment illustrates how the present invention improves the ability to identify slippage trends in real engineering slope scenarios through feature extraction and machine learning model training, demonstrating good recognition performance and scalability.

[0159] Example 3: Comparative Experiment of Slip Trend Prediction Model and Demonstration of Mobile Feedback Mechanism:

[0160] To further verify the superiority of the slip trend prediction model proposed in this invention in early warning of open-pit mine slopes, a set of comparative experiments were carried out, and a multi-terminal information feedback mechanism was developed to achieve multi-level response in the critical evolution stage of slope deformation.

[0161] I. Model Comparison Experiment Design:

[0162] We selected slope monitoring data from a large open-pit copper mine in southern China, including stimulus-response data and artificial stability labels for two consecutive months (approximately 8,640 samples). The data covers the entire process of slope evolution from a stable state to the early stage of slippage.

[0163] The following three models were used to calculate the slip trend score:

[0164] Model A (Traditional Statistical Threshold Method): Using the mean square response amplitude change rate as the judgment basis, an empirical threshold is set for early warning;

[0165] Model B (Support Vector Machine SVM): Employs radial basis kernel function to fit the boundary based on training samples;

[0166] Model C (the Random Forest (RF) model used in this invention): as shown in the training process in Example 2;

[0167] The performance of the three models was compared using 10-fold cross-validation on the same validation set. The main evaluation metrics are shown in Table 2 below:

[0168]

[0169] Table 2

[0170] As shown in Table 2, the results indicate that the slip trend recognition model constructed in this invention outperforms traditional methods in multiple indicators such as accuracy, false alarm rate, and false negative rate, demonstrating significant intelligent discrimination and generalization capabilities.

[0171] This invention aims to address the problems existing in current open-pit mine slope stability monitoring and early warning methods, such as response lag, insufficient early warning lead time, difficulty in quantifying mechanical evolution, and weak intelligent analysis capabilities. It constructs a slope slip early warning system with active excitation capability, continuous response monitoring, and intelligent trend recognition, so as to achieve early identification and graded response to the slope structure stability change process.

[0172] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A slope slip trend early warning system for open-pit mines based on active micro-vibration excitation, characterized in that, Includes the following components: Excitation unit: Composed of several adjustable frequency micro-vibration exciters, which periodically apply low-amplitude controllable mechanical disturbance signals to the slope rock mass. , To stimulate the structural response of soil and rock media and improve early warning sensitivity; Sensing unit: includes a triaxial accelerometer Displacement gauge GNSS sensor Ground acoustic wave sensor Record vibration disturbance signals Response changes during propagation and data acquisition ; Analysis Unit: Includes an edge intelligent terminal, a slip trend recognition model built into the edge intelligent terminal, and a lightweight neural network. The slip trend recognition model analyzes the data. Perform sensor processing and analyze the data at each time step. Classify and output slip trend labels; Interactive unit: including command center platform, field mobile terminal and remote data management system, which receives and parses the early warning information output by slip trend recognition model and then executes response; Communication control unit: Used to ensure stable data flow and coordinated operation of the above units and modules; The main control module that controls the adjustable frequency micro-vibration exciter issues a unified timestamp command. A frequency-adjustable micro-vibration exciter initiates micro-vibration disturbances, and a triaxial accelerometer is used. Displacement gauge GNSS sensor Ground acoustic wave sensor exist to Data is collected synchronously within the time window, including: The edge intelligent terminal represents the response duration window. It performs preliminary filtering, threshold removal, and event marking on the collected signals, numbering the valid responses under each stimulus as follows: ,in The stimulus-response cycle sequence numbers are as follows: ； in, This represents the triaxial values ​​acquired by each acceleration channel; Represents the GNSS displacement response vector; This response data represents the change in angle. The data is then uploaded to an edge intelligent terminal as input features for the slip trend recognition model.

2. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation according to claim 1, characterized in that, The adjustable frequency micro-vibration exciter is anchored to the slope surface, slope toe, and stabilizing support to generate disturbance signals. The excitation controller generates the data in the form of a function, as follows: ； in, : Initial amplitude of the excitation; Excitation frequency; Initial phase; : Time variable.

3. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation according to claim 2, characterized in that, The triaxial accelerometer Displacement gauge GNSS sensor Ground acoustic wave sensor Generate original data vector , ,in, Triaxial acceleration; Displacement; GNSS plane offset; Ground acoustic signal response, all sensors use the same sampling frequency. The timestamp is synchronously distributed by the communication control unit.

4. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation according to claim 3, characterized in that, The lightweight neural network is ,definition , The fused feature vector, generated by the data fusion module, represents the slope slippage trend classification result: 0 = safe, 1 = weak anomaly, and 2 = strong anomaly.

5. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation according to claim 4, characterized in that, After the response data enters the edge intelligent terminal, it is necessary to perform synchronous fusion and unified formatting of multi-source sensor signals to construct a high-dimensional feature vector for slip trend recognition. The specific steps are as follows: S01: The edge intelligent terminal embeds a data fusion and feature extraction module, which is triggered by an incentive time. Based on this, the alignment, cleaning, and feature extraction of different physical quantity data within a time window are achieved; S02: Perform bandpass filtering and normalization preprocessing on the input data from each sensor: for acceleration response Eliminate background noise and low-frequency drift; for displacement data Apply the moving average to remove platform drift; S03: All preprocessed signals are sampled at a uniform frequency. Interpolation resampling is performed below, and in the window Inner alignment to form a uniform response matrix: ； in: For time series points, Indicates the first The multidimensional response fusion matrix corresponding to the secondary stimulus; S04: A composite feature set is constructed using three types of indices: time domain, frequency domain, and nonlinear dynamics. Each excitation response data is mapped to a feature vector of fixed length. ； in Indicates the first Each feature extraction function in the fusion matrix The result of the operation is the total feature dimension.

6. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation according to claim 5, characterized in that, The slippage trend recognition model uses feature vectors As input, output the slip trend level of the slope under its current condition. , respectively representing "stable state", "gradually changing state" and "critical glide state"; The method for constructing a slip trend recognition model is as follows: A01: A supplementary modeling approach using a random forest model as the primary identification framework, supplemented by a lightweight neural network for predicting slippage trend evolution. The random forest model is composed of... It consists of decision trees, each trained using random feature subsets and sample subsets, and the final prediction result is obtained by the voting integration of all trees; A02: Obtaining a slip trend recognition model: ; in, Indicates the first The decision function of each tree is mde{}, which is the mode function; each decision tree is divided into nodes based on the Gini index or entropy increase criterion, thereby constructing a layer-by-layer decision path for the slip state.

7. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation according to claim 6, characterized in that, Input feature vector of the slip trend recognition model and output slip trend level To train the model and build a training sample set: The total capacity of the sample library is Once the slip trend recognition model is deployed, it generates real-time predictive output after each round of excitation. A stability determination mechanism for the slip state was designed. When the slip state score exceeds the set threshold If this occurs, the system will automatically trigger an early warning mechanism.

8. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation according to claim 7, characterized in that, The early warning mechanism is as follows: The early warning mechanism uses the prediction results and state score indicators output by the slip state recognition model as its core trigger. The mechanism includes tiered response and multi-channel dissemination, where tiered response involves setting a state score. The lower limit of the threshold and Upper limit, according to The lower limit of the threshold and The upper limit achieves the corresponding response. When the system continuously collects m excitation responses, if the m consecutive responses are judged to the same high-level state, the early warning locking mechanism is activated, the early warning command is synchronized to the edge intelligent terminal and the interaction unit, and a digital early warning file is generated. The digital early warning file includes the excitation number, response characteristics, judgment level and slope code.

9. The open-pit mine slope slippage trend early warning system based on active micro-vibration excitation according to claim 8, characterized in that, The multi-channel release includes a local audible and visual alarm device, a control room display module, and a mobile client. The local audible and visual alarm device issues an audible and visual alarm in the slope operation area, the control room display module displays a real-time pop-up alarm on the GIS interface, and sends SMS or push notifications. The mobile client displays the warning simultaneously on a mobile app and web platform, enabling remote control.

Images