Underground coal machine working condition monitoring system and method based on operation state

Abstract

The invention discloses an underground coal machine working condition monitoring system and method based on a running state, particularly relates to the field of running monitoring of underground coal mine conveying equipment, and is used for solving the problems of working condition misjudgment and intervention failure caused by reference drift and unknown abnormal attribution. Continuous tracking of the operation attitude is realized in cooperation with an inertia trajectory, and the attitude drift and false alarm problems caused by roadway deformation are avoided; by comparing the attitude matrix with a historical steady-state template, tiny offset of the equipment is recognized in time, and a stable reference is provided for abnormal judgment; physical characteristics representing impact and fluctuation are introduced, a dominant abnormal source is identified by utilizing machine learning, and a traditional judgment mode depending on experience is broken through; the whole process is linked layer by layer from attitude tracking, offset recognition to abnormal attribution, and finally scheduling suggestions are output, so that linkage closing of working condition recognition and intervention decision is realized, and the judgment stability in a complex scene is improved.

Description

Technical Field

[0001] This invention relates to the field of operation monitoring of underground coal mine conveying equipment, and more specifically, to an underground coal mining machinery condition monitoring system and method based on operating status. Background Technology

[0002] During continuous advancement of the underground mining face, the scraper conveyor extends and retracts with the step distance of the support, and the drive power and the posture of the trough side change in real time. Maintenance personnel often rely on coordinate monitoring architecture to judge the risk of upward or downward movement. The existing "Scraper Conveyor Condition Monitoring Method and Device" (application number 202410591831.4) establishes a follow-up coordinate reference at the nearest point on the centerline of the roadway, and uses a flexible measuring rod to obtain the three-dimensional position of each trough section to describe the spatial morphology of the conveyor, hoping to complete the condition judgment without the need for additional anchoring.

[0003] The tunnel centerline is a virtual curve, not a fixed physical marker. Floor bulging, progressive rock compression, and support displacement errors collectively cause the centerline to drift continuously, resulting in a deviation in the coordinates calculated by the flexible measuring rod. When the simulated trajectory deviates from the actual position, the static threshold still judges it as upward or downward movement, easily triggering false alarms; conversely, if the drift direction cancels out the actual deformation, the alarm is delayed. Existing solutions lack absolute anchor points or inertial reference compensation links, creating blind spots for chronic tunnel deformation. Early warning reliability decreases with advancing distance, frequently interfering with scheduling decisions and impacting continuous mining efficiency.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an underground coal mining machine condition monitoring system and method based on operating status. By constructing a reference coordinate system with spatial anchoring capabilities, and cooperating with inertial trajectories, continuous tracking of operating attitude is achieved, avoiding attitude drift and false alarms caused by roadway deformation. Through comparison of the attitude matrix with historical steady-state templates, minute equipment offsets are identified in a timely manner, providing a stable reference for anomaly judgment. Physical characteristics representing impacts and fluctuations are introduced, and machine learning is used to identify the dominant anomaly source, breaking through the traditional experience-based judgment method. The overall process is seamlessly connected from attitude tracking and offset identification to anomaly attribution, ultimately outputting scheduling suggestions, achieving a closed-loop linkage between condition identification and intervention decision-making, improving the stability of judgment in complex scenarios, and thus solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The method for monitoring the operating conditions of underground coal mining machinery based on its operational status includes the following steps:

[0008] S1: Set up absolute anchor point groups at equal intervals on both sides of the scraper conveyor, and simultaneously collect inertial trajectory sets to construct static reference coordinate frames and complete initial calibration.

[0009] S2: Perform differential fusion of the absolute anchor point group coordinates and the inertial trajectory set according to the acquisition time sequence, and output the real-time attitude matrix;

[0010] S3: Compare the real-time attitude matrix with the historical steady-state template, calculate the deformation drift vector and locate the tilt segment number;

[0011] S4: Compare the deformation drift vector with the support step sequence and the driving power sequence segment by segment to extract key features characterizing the displacement impact amplitude and power fluctuation complexity. After machine learning, generate the abnormal driving dominance sequence and complete the source priority sorting.

[0012] S5: Construct a set of early warning instructions for operating conditions based on the abnormal driving dominance sequence, and push targeted intervention suggestions to the field dispatch terminal.

[0013] In a preferred embodiment, step S1 includes the following:

[0014] A series of pressure-resistant visual anchor points are embedded at fixed intervals on both sides of the scraper conveyor. The three-dimensional coordinates of each pressure-resistant visual anchor point are obtained using high-precision measuring instruments to form a coordinate dataset of the absolute anchor point group. Inertial measurement units are installed at the drive head, tail, and curved section of the scraper conveyor to collect acceleration and angular velocity data in real time. Displacement is calculated by time integration of acceleration, and attitude is calculated by time integration of angular velocity. The data are then integrated to form an inertial trajectory set.

[0015] In a preferred embodiment, step S1 further includes the following:

[0016] After the scraper conveyor is installed, initial calibration is performed. The transformation relationship between the coordinate dataset of the absolute anchor point group and the inertial trajectory set is calculated by an optimization algorithm. The dynamic data in the inertial trajectory set is mapped to the coordinate system of the absolute anchor point group to construct a static reference coordinate frame.

[0017] In a preferred embodiment, step S2 includes the following:

[0018] The coordinates of the absolute anchor point group are cleaned to remove outliers and duplicate records; the Kalman filter algorithm is applied to the inertial trajectory set to reduce noise interference; the time series of the inertial trajectory set is adjusted to match the coordinates of the absolute anchor point group through linear interpolation to generate a synchronized dataset.

[0019] In a preferred embodiment, step S2 further includes the following:

[0020] At each time point, the difference vector between the inertial trajectory set and the coordinates of the absolute anchor point set is calculated; the inertial trajectory set is corrected using the difference vector and the adaptive fusion coefficient to generate the real-time attitude matrix; the observation equation is constructed to represent the coordinates of the absolute anchor point set as a combination of the real-time attitude matrix and the correction parameters; the optimal correction parameters are solved using the least squares method; the real-time attitude matrix is ​​adjusted according to the optimal correction parameters to obtain the optimized real-time attitude matrix.

[0021] In a preferred embodiment, step S3 includes the following:

[0022] Obtain the real-time attitude matrix of the scraper conveyor, calculate the difference in displacement and attitude between the real-time attitude matrix and the historical steady-state template, and generate the deformation drift vector; divide the scraper conveyor into multiple sections, calculate the average deformation drift vector of each section, and calculate the drift intensity based on the norm of the displacement drift vector; compare the drift intensity of each section, and locate the section number with the largest drift intensity as the tendency section number.

[0023] In a preferred embodiment, step S3 further includes the following:

[0024] The historical steady-state template is a central attitude matrix generated by classifying historical attitude data under normal operating conditions of the scraper conveyor through cluster analysis.

[0025] In a preferred embodiment, step S4 includes the following:

[0026] The difference sequence between adjacent step distances is calculated from the strut step distance sequence. Wavelet packet decomposition is performed on the difference sequence to extract the average amplitude of the high-frequency envelope as the step distance impact pulsation amplitude. The first-order difference of the driving power sequence is performed, and the complexity of the difference sequence is calculated using the sample entropy algorithm as the power fluctuation entropy value. The deformation drift vector, step distance impact pulsation amplitude, and power fluctuation entropy value are used as input features. The gradient boosting tree model is trained to output the abnormal driving dominance sequence. The abnormal driving dominance sequence is sorted in descending order to determine the priority of the abnormal source.

[0027] In a preferred embodiment, step S5 includes the following:

[0028] The system constructs a set of work status warning instructions by filtering the dominant sequence of anomalies to identify anomaly sources whose dominant values ​​exceed predefined thresholds. For each identified anomaly source, it generates a specific warning instruction containing an anomaly source identifier, anomaly type, and recommended intervention action. Targeted intervention suggestions are generated by evaluating the resource requirements of each warning instruction, comparing these requirements with the available resources at the field dispatch center, allocating resources based on the minimum of the required and available resources, and formulating intervention suggestions that include intervention actions, allocated resources, and priority levels. The intervention suggestions are then pushed to the field dispatch center by encoding the set of intervention suggestions into data packets, transmitting them via a wireless communication network, and having the field dispatch center decode the data packets and present them to the dispatchers for execution.

[0029] The underground coal mining machinery condition monitoring system based on operational status includes:

[0030] Anchoring framing unit: Absolute anchor point groups are embedded at equal intervals on both sides of the scraper conveyor, and inertial trajectory sets are collected simultaneously to construct a static reference coordinate frame and complete the initial calibration.

[0031] Fusion attitude solution unit: performs differential fusion of the absolute anchor point set coordinates and the inertial trajectory set according to the acquisition time sequence, and outputs the real-time attitude matrix;

[0032] Offset identification unit: Compares the real-time attitude matrix with the historical steady-state template, calculates the deformation drift vector, and locates the tilt segment number;

[0033] The main cause prioritization unit compares the deformation drift vector with the support step distance sequence and the driving power sequence segment by segment to extract key features characterizing the displacement impact amplitude and power fluctuation complexity. After machine learning, it generates an abnormal driving dominance sequence and completes the source priority ranking.

[0034] Early warning and suggestion unit: Constructs a set of early warning instructions for operating conditions based on the abnormal driving dominance sequence, and pushes targeted intervention suggestions to the field dispatch terminal.

[0035] The technical effects and advantages of the underground coal mining machinery condition monitoring system and method based on operational status of this invention are as follows:

[0036] This invention eliminates attitude drift caused by tunnel deformation during conveyor operation by constructing a reference coordinate system with spatial anchoring capabilities, solving the problems of false alarms and missed alarms caused by reference blurring in existing methods. Simultaneously, by combining inertial trajectory data, continuous tracking of equipment operating attitude is achieved. Comparison of the attitude matrix with historical steady-state templates allows for timely identification of minor equipment deviations, providing a reliable reference for anomaly judgment. Furthermore, two physical characteristic parameters representing mechanical impact and load fluctuations are introduced. Based on this, a machine learning model is used to comprehensively identify the most dominant anomaly sources, transforming the traditional experience-based judgment process into a data-driven decision output. Data flows between each step ensure that the judgment results are based on the continuous logic of dynamic tracking, deviation identification, and source ranking. Finally, the judgment results are transformed into scheduling suggestions fed back to the field, achieving closed-loop linkage between equipment condition identification, anomaly attribution, and intervention suggestions, significantly enhancing the stability and discrimination capability of the monitoring system in complex operating environments. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the method for monitoring the operating conditions of underground coal mining machinery based on operational status, as described in this invention.

[0038] Figure 2 This is a schematic diagram of the underground coal mining machinery condition monitoring system based on the operating status of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example 1: Figure 1 This invention presents a method for monitoring the operating conditions of underground coal mining machinery based on its operational status, comprising:

[0041] S1: Set up absolute anchor points at equal intervals on both sides of the scraper conveyor, and simultaneously collect inertial trajectory sets to construct a static reference coordinate frame and complete the initial calibration.

[0042] S2: Perform differential fusion of the absolute anchor point set coordinates and the inertial trajectory set according to the acquisition time sequence to output the real-time attitude matrix.

[0043] S3: Compare the real-time attitude matrix with the historical steady-state template, calculate the deformation drift vector, and locate the tendency segment number.

[0044] S4: Compare the deformation drift vector with the support step sequence and the driving power sequence segment by segment to extract key features characterizing the displacement impact amplitude and power fluctuation complexity. After machine learning, generate the abnormal driving dominance sequence and complete the source priority sorting.

[0045] S5: Construct a set of early warning instructions for operating conditions based on the abnormal driving dominance sequence, and push targeted intervention suggestions to the field dispatch terminal.

[0046] In underground coal mining, scraper conveyors, as core transportation equipment, are crucial to production efficiency and safety. The underground roadway environment is complex and variable. Affected by factors such as floor bulging, surrounding rock compression, and support displacement errors, the roadway geometry undergoes slow deformation, leading to conveyor attitude drift. Existing monitoring methods rely on a virtual roadway centerline as a reference, but this centerline shifts with roadway deformation, resulting in unstable monitoring results, frequent false alarms or missed alarms, and impacting continuous mining efficiency. To address this issue, this invention proposes an underground coal mining machine condition monitoring method based on operational status. Step S1 introduces an absolute anchor point group and an inertial trajectory set to construct a stable static reference coordinate frame, laying a reliable foundation for subsequent attitude monitoring. This step aims to solve the reference drift problem caused by the lack of a fixed reference in traditional methods, ensuring the stability of the monitoring coordinate system and providing a reliable spatial reference for subsequent steps.

[0047] The goal of step S1 is to embed absolute anchor point groups at equal intervals on both sides of the scraper conveyor, and simultaneously collect inertial trajectory sets to construct a static reference coordinate frame and complete the initial calibration.

[0048] 1-1: Setting up the absolute anchor point group;

[0049] On the sides of the scraper conveyor, a series of pressure-resistant visible anchor points are embedded at fixed intervals (e.g., every 5 meters). These anchor points are made of high-strength materials to ensure their stability under the pressure of the surrounding rock in the roadway, and are visually marked for easy identification in subsequent measurements. Using high-precision measuring instruments, such as a total station, each anchor point is located, and its three-dimensional coordinates in the roadway coordinate system are obtained, including longitudinal, lateral, and vertical height information. These coordinates are compiled and recorded into a set, forming a coordinate dataset of the absolute anchor point group.

[0050] The specific processing procedure is as follows: First, determine the installation positions of the anchor points on the roadway sidewall. Starting from the beginning of the roadway, mark a point at fixed intervals along the length of the conveyor to ensure that the anchor points are evenly distributed and cover the entire length of the conveyor. Next, use drilling equipment to embed the anchor points at the marked positions. The anchor point material must have high compressive strength, such as steel or composite material anchor rods, designed to withstand the force applied by the surrounding rock without displacement. After installation, start a high-precision measuring instrument to scan and locate each anchor point sequentially. For each anchor point, the measuring instrument calculates the distance and angle of the anchor point relative to the instrument's reference point by emitting a laser beam and receiving the reflected signal. Then, combined with the known position of the instrument in the roadway coordinate system, the three-dimensional coordinate value of the anchor point is calculated. The calculation process is as follows: taking the instrument position as the origin, the offset of the anchor point in the longitudinal, transverse, and vertical directions is determined based on the distance and angle using trigonometric functions. Then, the instrument's own coordinate value is added to obtain the absolute position of the anchor point. The coordinate values ​​of all anchor points are finally summarized into an ordered coordinate dataset for subsequent spatial reference.

[0051] Because the sidewalls are solid structures on both sides of the tunnel, they are less affected by deformation compared to the floor or roof, making them suitable as anchor points. The uniform distribution and compressive strength design of the anchor points ensure that the entire system maintains a stable spatial reference point even when the tunnel environment changes. In this way, the monitoring system obtains a fixed reference frame, effectively addressing the positional drift caused by tunnel deformation and providing reliable spatial data for subsequent attitude monitoring.

[0052] 1-2: Acquisition of inertial trajectory sets;

[0053] Inertial measurement units (IMUs) are installed at key locations on the scraper conveyor, such as the drive head, tail, and curved sections. Each IMU contains a three-axis accelerometer and a three-axis gyroscope, used to measure the acceleration and angular velocity of the equipment in three directions, respectively. The IMUs acquire data in real time at a fixed frequency. By processing the acquired acceleration and angular velocity data, the displacement and attitude information of the conveyor are calculated, thus forming a dynamic trajectory data set, called the inertial trajectory set.

[0054] The specific processing procedure is as follows: Before the conveyor starts operating, the inertial measurement unit is fixed to key locations to ensure its synchronous movement with the equipment. During operation, a three-axis accelerometer records the conveyor's acceleration values ​​in the longitudinal, transverse, and vertical directions, while a three-axis gyroscope records the angular velocity values ​​around the three axes. The collected data is stored in time-series format. First, the acceleration data is processed by performing two consecutive time integrations on the acceleration over time to calculate the equipment's displacement in each direction.

[0055] Specifically, for acceleration in a certain direction, with the initial velocity known or assumed to be zero, the change in velocity is obtained through a first integration, which is the sum of the product of acceleration and time in each time interval. A second integration of the velocity yields the change in displacement, also a sum of the product of velocity and time, ultimately determining the displacement distance of the equipment in that direction. Simultaneously, the angular velocity data is processed. By performing a time integration on the change in angular velocity over time, the rotation angle of the equipment around each axis is calculated, resulting in the sum of the product of angular velocity and time in each time interval, thus obtaining the change in attitude. The displacement and attitude data of all key components are then integrated in chronological order to form a dynamic trajectory dataset containing both position and angle information.

[0056] Inertial measurement units (IMUs) are a core tool for capturing the dynamic changes of conveyors because they can record the motion state of equipment in real time, and are especially suitable for scenarios that require monitoring attitude adjustments during operation. The combination of collected dynamic data and static anchor point data can comprehensively reflect the operating status of the conveyor, providing rich dynamic information support for subsequent analysis and ensuring that the monitoring system is not limited to static position descriptions.

[0057] 1-3: Construct a static reference coordinate frame and complete the initial calibration;

[0058] After the scraper conveyor is installed, a comprehensive attitude calibration is performed. The coordinate dataset of the absolute anchor point group is correlated with the inertial trajectory set, and the transformation relationship between the two is calculated. This ensures that the dynamic data in the inertial trajectory set can be mapped to the coordinate system of the absolute anchor point group. The transformation relationship includes rotation and translation components, which are solved using an optimization algorithm to ensure that the mapped data is highly consistent with the anchor point coordinates.

[0059] The specific processing procedure is as follows: First, with the conveyor stationary, the initial displacement and attitude data of the inertial measurement unit (IMU) are recorded as the starting point of the inertial trajectory set. Simultaneously, the coordinates of several anchor points closest to the IMU's position in the absolute anchor point group are selected as reference points. Next, the spatial relationship between the starting point of the inertial trajectory set and the coordinates of the absolute anchor point group is calculated. Assuming the initial displacement record of an IMU is a three-dimensional position point, the goal is to adjust this position point to match the coordinates of the nearest anchor point. To achieve this, a rotation relationship and a translation distance need to be determined. The rotation relationship is calculated based on the directional difference between the two sets of coordinates. By comparing the attitude angle recorded by the IMU with the directional reference of the anchor point coordinate system, an angle adjustment amount in three-dimensional space is calculated to align the two directions. The translation distance is calculated based on the positional difference between the two sets of coordinates. By subtracting the difference between the IMU's position and the nearest anchor point's position, a three-dimensional offset is obtained. Combining rotation and translation, a transformation relationship is formed to map all subsequent data points in the inertial trajectory set to the coordinate system of the absolute anchor point group. To improve accuracy, optimization methods, such as the least squares method, are employed. Specifically, the rotation and translation parameters are adjusted with the goal of minimizing the sum of squared coordinate deviations between all key components and their corresponding anchor points. This process is iterated repeatedly until the deviations converge to a minimum, ultimately determining the transformation relationship. After mapping, the dynamic data in the inertial trajectory set and the static coordinates of the absolute anchor point set are unified.

[0060] The initial calibration eliminated data inconsistencies caused by installation errors or differences in the initial state of the equipment, ensuring consistency between dynamic data and the static baseline. The constructed static baseline coordinate frame integrates the two data sources into a unified reference system, enabling the monitoring system to accurately describe the spatial state of the conveyor and providing precise spatial positioning data for subsequent analysis.

[0061] By embedding a set of absolute anchor points in the roadway walls and collecting inertial trajectory sets, a static reference coordinate frame for monitoring the scraper conveyor was successfully constructed, solving the problem of reference instability caused by roadway centerline drift. The absolute anchor point set provides fixed spatial reference points through high-precision positioning, while the inertial trajectory set reflects the conveyor's operating status through dynamic data acquisition. Initial calibration achieves spatial unification of the two through calculation of transformation relationships. This combination of static and dynamic methods enables the monitoring system to maintain coordinate system stability even in complex roadway environments, ensuring the reliability and consistency of spatial data and providing a solid spatial reference foundation for subsequent monitoring tasks.

[0062] Step S1 establishes a static reference coordinate frame and completes initial calibration by embedding absolute anchor point sets on both sides of the scraper conveyor and simultaneously acquiring inertial trajectory sets, laying the foundation for subsequent attitude monitoring. However, chronic deformation of the roadway environment (such as floor bulging, surrounding rock compression, and support displacement errors) can cause conveyor attitude drift. A single static anchor point set cannot capture dynamic changes, while the inertial trajectory set, although dynamic, is susceptible to noise and accumulated errors. Therefore, step S2 requires effectively fusing the absolute anchor point set coordinates with the inertial trajectory set to generate a real-time attitude matrix. This eliminates drift caused by roadway deformation, ensures the stability of the monitoring coordinate system, and provides reliable input for the deformation drift vector calculation in step S3.

[0063] The goal of step S2 is to perform differential fusion of the absolute anchor point set coordinates and the inertial trajectory set according to the acquisition time sequence, and output the real-time attitude matrix.

[0064] 2-1: Data preprocessing;

[0065] Before fusing the absolute anchor point coordinates and the inertial trajectory set, both types of data need to be preprocessed to ensure data quality and temporal consistency. The absolute anchor point coordinates are derived from high-precision measuring instruments and may contain outliers due to sensor malfunctions or roadway obstructions. The inertial trajectory set is derived from inertial measurement units and is susceptible to noise interference, affecting the accuracy of displacement and attitude calculations. Furthermore, the acquisition frequencies of these two types of data may differ, necessitating time synchronization to align the data points. The preprocessing process aims to provide high-quality input data for subsequent fusion through cleaning, filtering, and synchronization operations.

[0066] Preprocessing begins with cleaning the coordinates of the absolute anchor point group. The first step is to identify and remove outliers. This involves calculating the Euclidean distance between two adjacent anchor points and setting a reasonable distance threshold, such as 1.5 times the anchor point spacing. If the distance between an anchor point and the previous anchor point exceeds this threshold, the anchor point is considered an outlier and removed from the dataset. To calculate the Euclidean distance, the differences between the coordinate components of each anchor point in 3D space are taken, these differences are squared one by one, summed, and the square root of the result is taken to obtain the distance between the two points. The threshold is set based on the physical characteristics of the anchor point distribution to ensure that only points significantly deviating from the normal range are removed. The second step is to check if there are identical coordinate records in the dataset. If two records are found to have all their coordinate components equal in 3D space, the later record is deleted to eliminate duplicates. Through this cleaning process, outliers and redundant data in the absolute anchor point group coordinates are effectively removed, ensuring data reliability.

[0067] Next, the inertial trajectory set is filtered to reduce the impact of noise on acceleration and angular velocity data. A Kalman filter algorithm is used here, optimizing the state estimation through two stages: prediction and update. First, in the prediction stage, the predicted state at the current moment is calculated using the state estimate (including displacement and attitude) from the previous moment, combined with the system dynamics model. The system dynamics model describes the changes in state quantities over time, typically based on the integral relationship between acceleration and angular velocity. The predicted state is an extrapolation based on existing data. Then, in the update stage, the predicted state is corrected using the actual measured values ​​at the current moment (i.e., the acceleration and angular velocity output by the inertial measurement unit). Specifically, the difference between the predicted state and the measured value is adjusted according to certain weights to obtain the optimal estimate for the current moment. The determination of the weights depends on the estimation of the noise covariance, ensuring a balance between prediction and measurement in the filtering result. After Kalman filtering, noise in the inertial trajectory set is significantly suppressed, and the smoothness of the displacement and attitude data is improved.

[0068] Finally, time synchronization is performed to resolve the inconsistency in the acquisition frequency between the absolute anchor point coordinates and the inertial trajectory set. The goal of synchronization is to align the time series of the inertial trajectory set with the time points of the absolute anchor point coordinates. Specifically, at each time point of the absolute anchor point coordinates, linear interpolation is performed on the displacement and attitude data of the inertial trajectory set. The linear interpolation calculation process involves first finding the two existing time points in the inertial trajectory set that are closest to the target time point, denoted as the previous time point and the next time point, respectively; then, calculating the interval between the target time point and the previous time point, as well as the total interval between the previous and next time points; next, the difference in displacement or attitude values ​​corresponding to the two time points is proportionally allocated to the target time point to obtain the interpolation result. After interpolation, an inertial trajectory dataset consistent with the time series of the absolute anchor point coordinates is generated. Through time synchronization, the two types of data are unified in the time dimension, creating conditions for subsequent fusion.

[0069] Through the above cleaning, filtering and synchronization processes, outliers, noise and temporal differences in the absolute anchor point coordinates and inertial trajectory sets are effectively eliminated, the data quality is significantly improved, and a reliable input is provided for subsequent differential fusion.

[0070] 2-2: Differential fusion;

[0071] After data preprocessing, the differential fusion stage begins. This stage combines the stability of the absolute anchor point coordinates with the dynamics of the inertial trajectory set to generate a real-time attitude matrix. The absolute anchor point coordinates provide a static reference, correcting drift errors caused by noise accumulation in the inertial trajectory set; the inertial trajectory set reflects real-time changes in the device, capturing dynamic attitude information. Differential fusion eliminates drift errors and preserves dynamic characteristics by calculating and adjusting the differences between the two types of data.

[0072] The fusion process begins with the calculation of the difference vector. At each time point, the coordinates of the synchronized absolute anchor point group and the displacement data from the inertial trajectory set are extracted. The corresponding absolute anchor point group coordinates are subtracted from the displacement data of the inertial trajectory set to obtain the difference vector. Specifically, the difference vector is calculated in three-dimensional space by calculating the difference between the two sets of data at each coordinate component, forming a three-dimensional vector. The magnitude and direction of this vector reflect the deviation of the inertial trajectory set from the coordinates of the absolute anchor point group, i.e., the degree of drift. The introduction of the difference vector provides a quantitative basis for subsequent corrections.

[0073] Next, the inertial trajectory set is corrected using differential vectors to generate a real-time attitude matrix. The correction process involves subtracting the product of the fusion coefficient and the differential vector from the displacement data in the inertial trajectory set to obtain the corrected displacement data. The fusion coefficient is a value between 0 and 1, used to control the relative contribution of the absolute anchor point group coordinates and the inertial trajectory set in the fusion process. Specifically, each component of the differential vector is first multiplied by the fusion coefficient to obtain a weighted adjustment; then, this adjustment is subtracted component by component from the displacement data of the inertial trajectory set to generate new displacement data. The new displacement data, combined with the attitude information in the inertial trajectory set, forms the real-time attitude matrix. To adapt the fusion effect to different situations, the fusion coefficient adopts an adaptive adjustment method. The adjustment process involves calculating the change in the difference vector at adjacent time points, i.e., subtracting the difference vector from the previous time point from the current time point's difference vector to obtain the change vector. Then, the squares of each component of the change vector are summed, and the square root is taken to obtain the magnitude of the change. Next, this change is divided by the time interval between the two time points to obtain the rate of change. Finally, the rate of change is multiplied by a preset adjustment parameter to obtain the fusion coefficient. The adjustment parameter is determined based on system characteristics to ensure that the coefficient changes within a reasonable range. When the rate of change is large, the fusion coefficient increases, enhancing the influence of the absolute anchor point coordinates to suppress drift; when the rate of change is small, the fusion coefficient decreases, preserving more dynamic information from the inertial trajectory set.

[0074] Through differential fusion, the real-time attitude matrix achieves dynamic correction of drift error by fusing the stability of the absolute anchor point set coordinates and the dynamics of the inertial trajectory set, while maintaining sensitivity to changes in device attitude, providing reliable preliminary results for subsequent optimization.

[0075] 2-3: Attitude matrix optimization;

[0076] The real-time attitude matrix generated by differential fusion may still contain minor errors, requiring further optimization to improve accuracy. The optimization process uses the coordinates of the absolute anchor point group as observation constraints to adjust the real-time attitude matrix, making it more closely match the actual position. The optimization aims to enhance the accuracy and stability of attitude estimation through systematic error minimization.

[0077] Optimization begins with constructing the observation equations. The absolute anchor point coordinates are represented as a combination of the real-time attitude matrix and a correction parameter to be estimated, taking into account observation errors. The correction parameter is a spatial transformation matrix used to describe the adjustment amount of the real-time attitude matrix. Specifically, the observation equations are constructed as follows: at each time point, the absolute anchor point coordinates are considered as the result of the real-time attitude matrix acting on the initial position, then adjusted by the correction parameter; the deviation is represented by the observation error. The observation error originates from measurement noise and fusion residuals.

[0078] Subsequently, the least squares method was used to solve the optimization problem. The goal was to minimize the sum of squared observation errors at all time points. Specifically, the observation error at each time point was first calculated, which is the difference between the coordinates of the absolute anchor point group and the predicted value after adjusting the real-time attitude matrix. Then, the error components at all time points were squared and summed to obtain the total error. Next, the total error was repeatedly calculated by adjusting the values ​​of the correction parameters to find the correction parameters that minimized it. During the adjustment process, each element of the correction parameters was gradually changed until the total error converged to its minimum value. After the solution was obtained, the optimal correction parameters were obtained.

[0079] Finally, the real-time attitude matrix is ​​adjusted based on the optimal correction parameters. Specifically, the optimal correction parameters are applied to the real-time attitude matrix to transform its displacement and attitude components. The transformation process involves adding the displacement data of the real-time attitude matrix to the translation component in the correction parameters, and multiplying the attitude data to the rotation component in the correction parameters to generate the optimized real-time attitude matrix. The optimized real-time attitude matrix has a position and orientation in three-dimensional space that are closer to the actual values.

[0080] Through optimization, the minute errors in the real-time attitude matrix are effectively reduced, and the accuracy of attitude estimation is further improved, providing high-quality output data for subsequent applications.

[0081] Through three steps—data preprocessing, differential fusion, and attitude matrix optimization—the coordinates of the absolute anchor point group and the inertial trajectory set are effectively integrated, outputting an accurate real-time attitude matrix. Preprocessing eliminates outliers, noise, and temporal differences in the data; differential fusion combines static stability and dynamic characteristics to correct drift errors; and attitude matrix optimization further reduces residual errors through mathematical methods. This progressively refined method successfully addresses the challenges of attitude estimation in the complex environment of underground roadways, ensuring the reliability and accuracy of the output results.

[0082] Step S2 integrates the coordinates of the absolute anchor point group with the inertial trajectory set using differential fusion technology, outputting a real-time attitude matrix and eliminating dynamic drift errors. However, the real-time attitude matrix only reflects the current state of the scraper conveyor and cannot directly reveal whether there are abnormal deformations or potential risks. Therefore, step S3 aims to calculate the deformation drift vector and locate the tendency segment number by comparing the real-time attitude matrix with the historical steady-state template, providing a precise location basis for subsequent anomaly source identification.

[0083] The goal of step S3 is to calculate the deformation drift vector and locate the tendency segment number by comparing the real-time attitude matrix with the historical steady-state template, so as to identify the abnormal tendency segment in the scraper conveyor's operating attitude.

[0084] 3-1: Construction of historical steady-state templates;

[0085] Before analyzing whether the scraper conveyor's operating posture is abnormal, it is necessary to establish a normal state benchmark for comparison, namely a historical steady-state template. The historical steady-state template is generated by processing historical posture data recorded under normal operating conditions of the scraper conveyor, reflecting the typical posture characteristics of the equipment when no abnormalities occur. The following is the complete process for constructing the historical steady-state template.

[0086] First, historical attitude data recorded during normal operation of the scraper conveyor was collected. This data consists of real-time attitude matrices for multiple time periods. Each real-time attitude matrix contains the equipment's displacement and attitude information, where the displacement information represents the equipment's position coordinates in space, and the attitude information represents the equipment's orientation in space. To extract stable attitude patterns from this historical attitude data, cluster analysis was used to classify the data.

[0087] Cluster analysis groups similar attitude patterns together by comparing the similarity between real-time attitude matrices across different time periods. The specific steps are as follows: For any two real-time attitude matrices, calculate the difference between their displacement information. This involves taking the difference between each component of the displacement information of the two matrices, summing the squares of these differences, and taking the square root to obtain the distance between the displacement information. For attitude information, which is usually represented by quaternions, calculate the rotation angle between two quaternions by multiplying the inverse of one quaternion by the other to obtain a quaternion representing the difference in rotation. Then, extract the rotation angle from this quaternion. Combining the distance between displacement information and the rotation angle between attitude information forms a similarity metric. Real-time attitude matrices with high similarity are grouped together. After cluster analysis, several groups of attitude patterns are obtained, each representing a common normal attitude.

[0088] Within each attitude pattern group, a representative central attitude matrix is ​​calculated to serve as the typical attitude representation for that group. The central attitude matrix is ​​generated using a weighted average method. The specific steps are as follows: First, a weight value is determined for each attitude pattern group. This weight value is based on the frequency of occurrence or stability of that group's attitude pattern in historical attitude data; attitude patterns with higher frequency of occurrence or less variation receive higher weight values. Next, a weighted average is calculated to generate the central attitude matrix. For displacement information, each component of the displacement information in each real-time attitude matrix within the same group is multiplied by its corresponding weight value. All weighted components are summed, and then divided by the sum of all weight values ​​to obtain the displacement information components of the central attitude matrix. For attitude information, since attitude is represented by quaternions, direct weighted averaging is complex. Therefore, each quaternion is first converted into a corresponding rotation matrix. Then, all rotation matrices are weighted according to their weight values; that is, each element of each rotation matrix is ​​multiplied by its corresponding weight value, summed, and then divided by the sum of the weight values ​​to obtain the average rotation matrix. Finally, this average rotation matrix is ​​converted back to a quaternion to serve as the attitude information for the central attitude matrix. After calculating the center attitude matrix for each attitude pattern, the center attitude matrices of all groups are integrated to form a historical steady-state template. The historical steady-state template can contain one or more center attitude matrices, the specific number depending on the actual application requirements.

[0089] Through the above cluster analysis and weighted average processing, the historical steady-state template can accurately reflect the typical posture characteristics of the scraper conveyor during normal operation, providing an accurate reference standard for subsequent judgment of whether the posture is abnormal.

[0090] 3-2: Calculation of deformation drift vector;

[0091] After generating the historical steady-state template, it is necessary to quantify the difference between the current attitude of the scraper conveyor and its attitude under normal conditions to determine whether the current attitude deviates from the normal range. This difference is represented in the form of a deformation drift vector, which contains deviation information in both displacement and attitude. The following is the specific calculation process of the deformation drift vector.

[0092] The calculation of the deformation drift vector begins with obtaining the real-time attitude matrix and the historical steady-state template. The real-time attitude matrix is ​​recorded by the system in real time and contains the displacement and attitude information of the scraper conveyor at a certain point in time; the historical steady-state template is generated by the aforementioned sub-step 1 and represents the attitude under normal conditions. First, the deviation of the displacement information, i.e., the displacement drift vector, is calculated. The specific steps are as follows: subtract each component of the displacement information of the real-time attitude matrix from each component of the displacement information of the corresponding center attitude matrix of the historical steady-state template to obtain a three-dimensional vector. Each component represents the positional offset of the equipment on the corresponding coordinate axis. This vector is the displacement drift vector, reflecting the deviation of the equipment in spatial position.

[0093] Next, the deviation of attitude information, i.e., attitude drift, is calculated. Since attitude information is represented by quaternions, attitude drift is determined by calculating the difference between the attitude quaternion of the real-time attitude matrix and the attitude quaternion of the central attitude matrix of the historical steady-state template. The specific steps are as follows: First, the inverse quaternion of the attitude quaternion of the historical steady-state template is calculated. Each component of the inverse quaternion is determined by the conjugate form of the original quaternion, i.e., the real part remains unchanged, and the imaginary part is the opposite. Then, this inverse quaternion is multiplied by the attitude quaternion of the real-time attitude matrix. The multiplication is performed according to the rules of quaternion arithmetic, i.e., the product of each component of the real and imaginary parts is calculated separately and like terms are combined to obtain a new quaternion, called the difference quaternion. This difference quaternion represents the rotational change required to rotate from the attitude of the historical steady-state template to the attitude of the real-time attitude matrix, reflecting the attitude deviation.

[0094] Finally, the displacement drift vector and the difference quaternion are combined to form the deformation drift vector. As a composite data structure, the deformation drift vector contains a three-dimensional displacement drift vector and a difference quaternion representing rotational changes, comprehensively describing the deviation characteristics of the scraper conveyor's current posture relative to its normal state.

[0095] Through the above calculation process, the deformation drift vector can accurately capture the attitude deviation of the scraper conveyor during operation, providing key quantitative data support for subsequent identification of abnormal areas.

[0096] 3-3: Positioning tendency segment number;

[0097] After calculating the deformation drift vector, it is necessary to analyze the degree of attitude deviation in different parts of the scraper conveyor to determine the section with the most significant attitude change, thereby identifying possible abnormal areas. The following is the complete processing procedure for the positioning tendency section number.

[0098] First, the scraper conveyor is divided into multiple sections, each assigned a number starting from 1. The total number of sections is determined based on the division method. Section division can be based on the physical structure of the equipment, such as dividing it into a drive head, tail section, and curved sections, or by uniformly dividing it according to a fixed length. Within each section, the average deformation drift vector is calculated to characterize its overall attitude deviation. The specific steps are as follows: For a given section, collect the deformation drift vectors at all time points within that section; for the displacement drift vector, sum the displacement drift vectors at all time points according to their components, that is, accumulate the values ​​of each component one by one, and then divide by the total number of time points within that section to obtain the components of the average displacement drift vector; for the attitude drift difference quaternion, first convert each difference quaternion into a rotation matrix, then sum the corresponding elements of all rotation matrices and divide by the total number of time points to obtain the average rotation matrix, and then convert the average rotation matrix back into a quaternion as the attitude component of the average deformation drift vector.

[0099] Next, the drift intensity of each segment is calculated to quantify the degree of attitude deviation. The drift intensity is expressed as a scalar, and the specific steps are as follows: Square each component of the displacement drift vector in the average deformation drift vector; sum these squared values ​​and take the square root to obtain a numerical value in meters. This value represents the drift intensity, reflecting the average magnitude of the spatial deviation of that segment. The difference quaternion of attitude drift can be retained at this stage for subsequent analysis, but the calculation of drift intensity is primarily based on displacement deviation.

[0100] Finally, compare the drift intensity of all segments to determine the segment number with the most significant attitude change. The specific steps are as follows: iterate through the drift intensity values ​​of each segment, record the largest value, and find the segment number corresponding to that value. This number is the tendency segment number, indicating that the attitude deviation of this segment is the highest and it may be the area where the anomaly occurred.

[0101] Through the above segment division and drift intensity analysis, the tendency segment number can be quickly and accurately determined, providing a precise positioning basis for identifying key abnormal areas in the scraper conveyor's operating posture.

[0102] By constructing a historical steady-state template, calculating the deformation drift vector, and identifying the segment sequence number of the orientation deviation, the identification of the most significantly changing segments in the operating attitude of the scraper conveyor was completed. The historical steady-state template generates an attitude baseline under normal conditions through clustering and weighted averaging of historical data; the deformation drift vector quantifies the deviation of the current attitude by calculating the difference between displacement and attitude; and the identification of the orientation deviation segment sequence number determines the key areas of attitude anomalies through segment analysis and drift intensity comparison. This method can effectively monitor the attitude changes of the scraper conveyor in complex environments, ensuring the accuracy and reliability of anomaly identification.

[0103] Step S3 calculates the deformation drift vector and locates the tendency segment number by comparing the real-time attitude matrix with the historical steady-state template. These steps lay the spatial reference and deformation quantification foundation for monitoring the conveyor's operating status. However, the deformation drift vector only reflects the amplitude and location of attitude changes and does not reveal the root cause of the anomaly. Therefore, step S4 needs to further analyze the relationship between the deformation drift vector and the support step distance sequence and drive power sequence, extract key features, and generate an anomaly drive dominance sequence to achieve accurate identification and priority ranking of anomaly sources, providing a basis for subsequent intervention.

[0104] The goal of step S4 is to extract key features characterizing the displacement impact amplitude and power fluctuation complexity by comparing the deformation drift vector with the support step distance sequence and the driving power sequence, apply machine learning to generate an abnormal driving dominance sequence, and complete the source priority ranking.

[0105] 4-1: When analyzing the operating status of a scraper conveyor, the support step distance sequence and drive power sequence are core data reflecting the dynamic behavior of the equipment. The support step distance sequence records the change in the step length of the equipment's advancement in the roadway, while the drive power sequence reflects the fluctuation of the equipment's load. To quantify the characteristics of step distance abrupt changes and power fluctuations, it is necessary to calculate the step distance impact pulsation amplitude and power fluctuation entropy value as characteristic criteria for identifying abnormal drive sources. The calculation process for both is described in detail below.

[0106] 4-1-1: Calculation process of step-width impact pulsation amplitude;

[0107] First, the differences between adjacent step distances are extracted from the stent step distance sequence to generate a step distance difference sequence. This sequence is obtained by subtracting adjacent step distances one by one, reflecting the amplitude and frequency of step distance changes. A large difference indicates a significant abrupt change in step distance at that point. Next, wavelet packet decomposition is applied to the step distance difference sequence. Wavelet packet decomposition separates high-frequency components containing abrupt changes by decomposing the sequence into multiple frequency sub-bands. Specifically, wavelet packet decomposition decomposes the step distance difference sequence layer by layer into low-frequency and high-frequency components, further decomposing the high-frequency component at each layer, ultimately forming sub-signals in multiple frequency ranges. High-frequency sub-signals are chosen here because they typically carry the instantaneous impact information of step distance abrupt changes. Subsequently, envelope signals are extracted from the high-frequency sub-signals. The envelope signal generation process involves smoothing the amplitude of the high-frequency sub-signals. Specifically, the absolute value of the high-frequency sub-signals is first taken, and then the amplitude variation is smoothed using a low-pass filter to obtain an amplitude curve that changes over time. This curve characterizes the intensity fluctuations of the high-frequency components. Finally, the average amplitude of the envelope signal is calculated. The average amplitude is calculated by summing the absolute values ​​of the envelope signal at all sampling points and then dividing by the total number of sampling points to obtain a numerical value in meters, called the step-size impact pulsation amplitude. The step-size impact pulsation amplitude quantifies the impact intensity during step-size changes, providing characteristic support for subsequent analysis in the displacement dimension.

[0108] 4-1-2: The calculation process of power fluctuation entropy;

[0109] For the driving power sequence, first-order differencing is performed to generate a power difference sequence. The power difference sequence is obtained by subtracting adjacent power values ​​in the sequence, reflecting the power's trend over time. A large difference indicates a significant power fluctuation at that point. Next, the sample entropy algorithm is applied to the power difference sequence to calculate its complexity. Sample entropy characterizes the disorder of power fluctuations by analyzing the degree of disorder in the sequence. Its calculation process is as follows: First, the power difference sequence is divided into multiple subsequences of fixed length, each containing several consecutive sampling points. Then, the similarity between these subsequences is compared one by one. Similarity is determined based on whether the difference between corresponding points in the subsequence is less than a preset threshold. The number of subsequence pairs satisfying the similarity condition is counted, and the logarithmic entropy value is calculated based on the statistical results of subsequences of different lengths. Specifically, sample entropy is obtained by comparing the ratio of the number of subsequence pairs of a certain length to the number of subsequence pairs after increasing the length by one, taking the natural logarithm to obtain a dimensionless value. The larger the power fluctuation entropy value, the more disordered the power difference sequence's fluctuations. Through this calculation, the power fluctuation entropy value quantifies the fluctuation characteristics of the driving power from the load dimension, providing a mathematical description for the identification of anomaly sources.

[0110] Through the above process, the step-pitch impact pulsation amplitude and power fluctuation entropy value respectively characterize the operating characteristics of the scraper conveyor from the perspectives of displacement abrupt change and load disorder, ensuring the comprehensiveness and accuracy of feature extraction.

[0111] 4-2: Detailed process of machine learning model training and anomaly-driven dominance generation;

[0112] Deformation drift vector, step-pitch impact pulsation amplitude, and power fluctuation entropy value describe the operating state of the scraper conveyor from three dimensions: attitude, displacement, and load, respectively. To comprehensively analyze the impact of these characteristics on the anomaly driving source, a gradient boosting tree model needs to be constructed, and an anomaly driving dominance sequence needs to be generated through nonlinear learning. The following details the model training and dominance generation process.

[0113] 4-2-1: Model training process;

[0114] First, the deformation drift vector, step-size impact pulsation amplitude, and power fluctuation entropy are used as input features to form the training dataset. The deformation drift vector originates from the quantification of equipment attitude changes in the preceding steps. These features together constitute a multi-dimensional input describing the equipment's operating state. The gradient boosting tree model makes predictions by integrating multiple decision trees. Each decision tree segments the data based on feature values ​​and outputs a preliminary result. The outputs of all trees are summed to form the final prediction. The specific training process includes: initializing a simple decision tree to minimize the error between its predicted and actual values; then, generating subsequent decision trees sequentially, with each new tree reducing the error by fitting the residuals of the previous prediction; finally, the model's prediction result is obtained by weighted summation of the outputs of all trees. To optimize model performance, parameters such as the number of trees, depth, and learning rate are iteratively adjusted during training. Cross-validation is also employed, dividing the dataset into multiple subsets. The model is trained on a subset and validated on the remaining subset in turn to ensure the model's predictive stability across different data. After training, the gradient boosting tree model can predict the impact of anomaly drivers based on the input features.

[0115] 4-2-2: The process of generating dominance driven by anomalies;

[0116] The trained gradient boosting tree model is applied to the complete dataset. Inputting sequences of deformation drift vectors, step-size impact pulsation amplitudes, and power fluctuation entropy values, the model outputs an anomaly-driven dominance sequence. Anomaly-driven dominance is a dimensionless value representing the degree of influence of each anomaly source on the conveyor's operating state. The specific generation process is as follows: for each set of input features, the model calculates a value through layer-by-layer splitting of the decision tree and output accumulation; this process is repeated for the entire dataset to generate a dominance sequence with the same length as the input sequence. A larger dominance value indicates a more significant impact of the corresponding anomaly source on the equipment's operating state. By analyzing the complex interactions between features, the gradient boosting tree model can accurately quantify the impact of each anomaly source, providing data support for subsequent processing.

[0117] Through the above process, the gradient boosting tree model achieves comprehensive analysis of multi-dimensional features, generates anomaly-driven dominance sequences, and ensures the accuracy and reliability of anomaly source impact assessment.

[0118] 4-3: Detailed process of source priority sorting;

[0119] The anomaly-driven dominance sequence reflects the degree of influence of each anomaly source on the operating status of the scraper conveyor. To facilitate the formulation of the implementation order of intervention measures, the anomaly-driven dominance sequence needs to be sorted to determine the priority of the anomaly sources. The sorting process is described in detail below.

[0120] 4-3-1: Sorting process;

[0121] First, obtain the anomaly-driven dominance sequence, which contains multiple values, each corresponding to the influence degree of an anomaly source. Next, sort these values ​​in descending order. The descending order process involves comparing the values ​​in the sequence and adjusting their positions one by one, arranging the values ​​from largest to smallest; if values ​​are equal, their relative order is maintained according to their original anomaly source numbers. After sorting, a new sequence is generated, where each value corresponds to an anomaly source and its dominance value. Based on this sequence, the priority of the anomaly sources is determined: the anomaly source with the largest dominance value is assigned the highest priority, decreasing sequentially until the anomaly source with the smallest dominance value is assigned the lowest priority. Finally, output the source priority ranking result, which includes the anomaly source number and its corresponding priority order.

[0122] Through the above process, the anomaly-driven dominance sequence is transformed into an ordered priority sequence, providing a clear sequential guide for the implementation of on-site intervention measures and ensuring the optimization of resource allocation and processing efficiency.

[0123] By calculating the step-pitch impact pulsation amplitude and power fluctuation entropy, key features characterizing displacement abrupt changes and load disorder were extracted. Using a gradient boosting tree model, the deformation drift vector, step-pitch impact pulsation amplitude, and power fluctuation entropy were comprehensively analyzed to generate an anomaly driving dominance sequence. By arranging the anomaly driving dominance sequence in descending order, the priority of the anomaly sources was determined. These steps, from feature extraction to model analysis and priority ranking, form a complete processing flow, linking attitude changes with the impact of anomaly sources. This ensures the accuracy and practicality of the monitoring method and provides system support for optimizing the operation status of scraper conveyors.

[0124] Step S4 extracts key features and generates an abnormal drive dominance sequence by analyzing the relationship between the deformation drift vector and the support step distance sequence and drive power sequence, thus completing the priority ranking of abnormal sources. These steps provide a complete link from benchmark construction to abnormal identification for monitoring the conveyor's operating status. However, the identification and ranking of abnormal sources only provide a basis for intervention and have not yet been directly transformed into on-site executable scheduling instructions. Therefore, step S5 needs to further construct a working condition early warning instruction set based on the abnormal drive dominance sequence and push targeted intervention suggestions to the on-site dispatching terminal, thereby transforming abnormal identification into closed-loop control intervention decisions and ensuring the practicality and efficiency of the monitoring method.

[0125] The goal of step S5 is to construct a set of early warning instructions for operating conditions through anomaly-driven dominance sequences and push targeted intervention suggestions to the field dispatch terminal.

[0126] 5-1: The complete process of constructing a working condition early warning instruction set;

[0127] The anomaly-driven dominance sequence contains multiple dimensionless values, each representing the degree of influence of an anomaly source on the system's operating state. To extract key anomaly sources requiring intervention from these values ​​and generate actionable instructions, a systematic screening and transformation process is needed to construct a set of operational condition warning instructions. The following is a detailed description of the specific processing procedure.

[0128] First, a preset threshold is set. This value is a dimensionless standard value derived from historical operational data analysis and is used to determine whether the impact of the anomaly source reaches a level requiring intervention. Each value in the anomaly driving dominance sequence is compared sequentially with this preset threshold. If a value is greater than the preset threshold, that value and its corresponding anomaly source are retained; if it is less than or equal to the preset threshold, it is excluded. Through this step-by-step comparison, a subset is selected from the anomaly driving dominance sequence. This subset contains all values ​​greater than the preset threshold and their corresponding anomaly sources. Each element in the subset consists of two parts: an identifier for the anomaly source, such as stent step distance anomaly or drive power anomaly; and the dominance value corresponding to that anomaly source.

[0129] Next, for each element in the subset, a specific early warning instruction is generated based on the type of its anomaly source. Each early warning instruction contains three parts: the anomaly source number, used to locate the specific anomaly; the anomaly type, used to clarify the nature of the problem; and suggested intervention measures, used to guide on-site operations. The generation of intervention measures is based on predefined rules for anomaly types. For example, when the anomaly source type is abnormal support step distance, the generated intervention measure is to adjust the support step distance to the normal range; when the anomaly source type is abnormal drive power, the generated intervention measure is to optimize the drive power settings. After the early warning instructions for all subset elements are generated, these instructions are integrated into a complete set, named the Operating Condition Early Warning Instruction Set. Each instruction in this set targets a specific anomaly source and clearly indicates the direction of intervention.

[0130] Through the above process, high-impact anomaly sources in the anomaly-driven dominance sequence are effectively identified and transformed into operational instructions. This screening and transformation method ensures that subsequent interventions can focus on key issues, improving the targeted nature of system operation state optimization.

[0131] 5-2: The complete process for generating targeted intervention recommendations;

[0132] The operational condition early warning command set provides early warning information for anomalies, but in actual operation, it is necessary to ensure that intervention measures match the on-site resource conditions. Therefore, it is necessary to assess the resource requirements of each early warning command and, in conjunction with available on-site resources, generate specific intervention recommendations. The following is a detailed description of the processing procedure.

[0133] First, for each early warning instruction in the operational condition early warning instruction set, determine the resource requirements for its execution. The resource type is determined based on the specific needs of the intervention measure, such as the time or manpower required to adjust the stent step distance. The resource quantity assessment is based on historical data or expert experience, and a specific value is preset for each intervention measure. For example, the time required to adjust the stent step distance is set to a certain number of seconds, and the manpower required is set to a certain number of people; the time required to optimize the drive power is set to a certain number of seconds. The resource requirements for each early warning instruction are determined by referring to these preset values ​​and recorded as the resource requirement quantity, with the unit consistent with the resource type.

[0134] Next, the total amount of resources currently available at the field dispatch terminal is obtained and recorded as resource availability, with its unit consistent with the resource demand. For example, if the resource type is time, then resource availability is the adjustable time that can be allocated on-site, in seconds. Then, for each warning instruction in the operational condition warning instruction set, the available resource allocation is calculated. The calculation method is as follows: compare the resource demand of each warning instruction with the on-site resource availability, and take the smaller value as the resource allocation. Specifically, if the resource demand of a warning instruction is less than or equal to the resource availability, then the resource allocation for that instruction is equal to its resource demand; if the resource demand is greater than the resource availability, then the resource allocation is equal to the resource availability. This method ensures that resource allocation does not exceed the on-site supply capacity.

[0135] After the resource allocation is determined, a targeted intervention suggestion is generated by combining the anomaly source information in each early warning instruction with the allocated resource amount. Each intervention suggestion includes three parts: specific intervention measures, such as adjusting the stent stride or optimizing the drive power; the allocated resource amount, such as the number of seconds or personnel; and the execution priority. The execution priority is determined based on the anomaly drive dominance value, with higher values ​​indicating higher priority. After all intervention suggestions for early warning instructions are generated, they are integrated into a set, named the Intervention Suggestion Set. Each suggestion in this set is an executable intervention plan.

[0136] Through the process of resource demand assessment and on-site resource matching, the intervention recommendation set ensures the feasibility of measures while achieving efficient resource utilization, providing clear guidance for on-site operations.

[0137] 5-3: The complete process for pushing intervention suggestions;

[0138] Once the intervention suggestion set is generated, it needs to be transmitted to the on-site dispatch terminal in a timely manner to support dispatchers in implementing intervention measures and realize the complete process from anomaly identification to execution. The following is a description of the specific processing procedure for pushing intervention suggestions.

[0139] First, each recommendation in the intervention recommendation set is converted into a digital format suitable for transmission, generating a data packet. The conversion process includes encoding information such as the anomaly source number, intervention description, resource allocation, and priority of each recommendation into a binary or hexadecimal numerical sequence. The data packet structure must conform to the requirements of the transmission protocol, including header information, data body, and checksum information, where the checksum information is used to verify data integrity. After encoding, the data packet contains the entire content of the intervention recommendation set.

[0140] Next, the data packets are transmitted to the receiving equipment at the field dispatch center via a wireless communication network. This wireless communication network must adapt to the complex conditions of the underground environment, supporting stable signal transmission and high transmission speeds. During transmission, the data packets are encapsulated according to the network protocol, and a checksum is added to ensure the receiving end can detect any data errors that may occur during transmission. After the data packets are transmitted from the sending end to the receiving end, the receiving equipment at the field dispatch center decodes them. The decoding process involves restoring the numerical sequence in the data packets to the original information of the intervention recommendation set, including the anomaly source number, intervention measures, resource allocation, and priority of each recommendation. During decoding, the integrity of the data is verified using the checksum; if an error is found, a retransmission is requested.

[0141] After decoding, the on-site dispatch center presents the reconstructed set of intervention recommendations to the dispatchers in a visual format, such as displaying detailed information for each recommendation on a screen or generating a printable report. The dispatchers then use this information to plan on-site actions, such as assigning personnel to adjust the support strut spacing or optimize drive power.

[0142] Through the processes of encoding, transmission, and decoding, the intervention recommendation set can be delivered to the field quickly and accurately, ensuring the timely implementation of intervention measures and achieving a closed-loop connection between monitoring and execution.

[0143] The three sub-steps described above together complete the entire process from anomaly identification to intervention execution. Constructing a condition early warning instruction set accurately identifies the anomaly sources requiring attention by filtering the anomaly-driven dominance sequence; generating targeted intervention suggestions ensures the operability of intervention measures through resource assessment and allocation; and pushing intervention suggestions ensures efficient implementation through data transmission and presentation.

[0144] Example 2: Figure 2 The present invention provides an underground coal mining machinery condition monitoring system based on operational status, comprising:

[0145] Anchoring framing unit: Absolute anchor point groups are embedded at equal intervals on both sides of the scraper conveyor, and inertial trajectory sets are collected simultaneously to construct a static reference coordinate frame and complete the initial calibration.

[0146] Fusion attitude solution unit: performs differential fusion of the absolute anchor point set coordinates and the inertial trajectory set according to the acquisition time sequence, and outputs the real-time attitude matrix;

[0147] Offset identification unit: Compares the real-time attitude matrix with the historical steady-state template, calculates the deformation drift vector, and locates the tilt segment number;

[0148] The main cause prioritization unit compares the deformation drift vector with the support step distance sequence and the driving power sequence segment by segment to extract key features characterizing the displacement impact amplitude and power fluctuation complexity. After machine learning, it generates an abnormal driving dominance sequence and completes the source priority ranking.

[0149] Early warning and suggestion unit: Constructs a set of early warning instructions for operating conditions based on the abnormal driving dominance sequence, and pushes targeted intervention suggestions to the field dispatch terminal.

[0150] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0151] It should be noted that the system of the present invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting a variety of hardware environments and usage requirements.

[0152] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0153] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely to distinguish one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0154] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for monitoring the operating conditions of underground coal mining machinery based on its operational status, characterized in that, Including the following steps: S1: Set up absolute anchor point groups at equal intervals on both sides of the scraper conveyor, and simultaneously collect inertial trajectory sets to construct static reference coordinate frames and complete initial calibration. S2: Perform differential fusion of the absolute anchor point group coordinates and the inertial trajectory set according to the acquisition time sequence, and output the real-time attitude matrix; S3: Compare the real-time attitude matrix with the historical steady-state template, calculate the deformation drift vector and locate the tilt segment number; S4: Compare the deformation drift vector with the support step sequence and the driving power sequence segment by segment to extract key features characterizing the displacement impact amplitude and power fluctuation complexity. After machine learning, generate the abnormal driving dominance sequence and complete the source priority sorting. S5: Construct a set of early warning instructions for operating conditions based on the abnormal driving dominance sequence, and push targeted intervention suggestions to the field dispatch terminal.

2. The method for monitoring the operating condition of underground coal mining machinery based on its operating status according to claim 1, characterized in that, Step S1 includes the following: A series of pressure-resistant visual anchor points are embedded at fixed intervals on both sides of the scraper conveyor. The three-dimensional coordinates of each pressure-resistant visual anchor point are obtained using high-precision measuring instruments to form a coordinate dataset of the absolute anchor point group. Inertial measurement units are installed at the drive head, tail, and curved section of the scraper conveyor to collect acceleration and angular velocity data in real time. Displacement is calculated by time integration of acceleration, and attitude is calculated by time integration of angular velocity. The data are then integrated to form an inertial trajectory set.

3. The method for monitoring the operating condition of underground coal mining machinery based on its operating status according to claim 2, characterized in that, Step S1 also includes the following: After the scraper conveyor is installed, initial calibration is performed. The transformation relationship between the coordinate dataset of the absolute anchor point group and the inertial trajectory set is calculated by an optimization algorithm. The dynamic data in the inertial trajectory set is mapped to the coordinate system of the absolute anchor point group to construct a static reference coordinate frame.

4. The method for monitoring the operating condition of underground coal mining machinery based on its operating status according to claim 3, characterized in that, Step S2 includes the following: The coordinates of the absolute anchor point group are cleaned to remove outliers and duplicate records; the Kalman filter algorithm is applied to the inertial trajectory set to reduce noise interference; the time series of the inertial trajectory set is adjusted to match the coordinates of the absolute anchor point group through linear interpolation to generate a synchronized dataset.

5. The method for monitoring the operating condition of underground coal mining machinery based on its operating status according to claim 4, characterized in that, Step S2 also includes the following: At each time point, calculate the difference vector between the coordinates of the inertial trajectory set and the coordinates of the absolute anchor point set; use the difference vector and adaptive fusion coefficients to correct the inertial trajectory set and generate the real-time attitude matrix; An observation equation is constructed, which represents the coordinates of the absolute anchor point group as a combination of the real-time attitude matrix and the correction parameters. The optimal correction parameters are solved using the least squares method. The real-time attitude matrix is ​​adjusted according to the optimal correction parameters to obtain the optimized real-time attitude matrix.

6. The method for monitoring the operating condition of underground coal mining machinery based on its operating status according to claim 5, characterized in that, Step S3 includes the following: Obtain the real-time attitude matrix of the scraper conveyor, calculate the difference in displacement and attitude between the real-time attitude matrix and the historical steady-state template, and generate the deformation drift vector; divide the scraper conveyor into multiple sections, calculate the average deformation drift vector of each section, and calculate the drift intensity based on the norm of the displacement drift vector; compare the drift intensity of each section, and locate the section number with the largest drift intensity as the tendency section number.

7. The method for monitoring the operating condition of underground coal mining machinery based on its operating status according to claim 6, characterized in that, Step S3 also Includes the following: The historical steady-state template is a central attitude matrix generated by classifying historical attitude data under normal operating conditions of the scraper conveyor through cluster analysis.

8. The method for monitoring the operating condition of underground coal mining machinery based on its operating status according to claim 6, characterized in that, Step S4 includes the following: The difference sequence between adjacent step distances is calculated from the strut step distance sequence. Wavelet packet decomposition is performed on the difference sequence to extract the average amplitude of the high-frequency envelope as the step distance impact pulsation amplitude. The first-order difference of the driving power sequence is performed, and the complexity of the difference sequence is calculated using the sample entropy algorithm as the power fluctuation entropy value. The deformation drift vector, step distance impact pulsation amplitude, and power fluctuation entropy value are used as input features. The gradient boosting tree model is trained to output the abnormal driving dominance sequence. The abnormal driving dominance sequence is sorted in descending order to determine the priority of the abnormal source.

9. The method for monitoring the operating condition of underground coal mining machinery based on its operating status according to claim 8, characterized in that, Step S5 includes the following: The system constructs a set of work status warning instructions by filtering the dominant sequence of anomalies to identify anomaly sources whose dominant values ​​exceed predefined thresholds. For each identified anomaly source, it generates a specific warning instruction containing an anomaly source identifier, anomaly type, and recommended intervention action. Targeted intervention suggestions are generated by evaluating the resource requirements of each warning instruction, comparing these requirements with the available resources at the field dispatch center, allocating resources based on the minimum of the required and available resources, and formulating intervention suggestions that include intervention actions, allocated resources, and priority levels. The intervention suggestions are then pushed to the field dispatch center by encoding the set of intervention suggestions into data packets, transmitting them via a wireless communication network, and having the field dispatch center decode the data packets and present them to the dispatchers for execution.

10. A monitoring system for the operating conditions of underground coal mining machinery based on its operational status, used to implement the monitoring method for the operating conditions of underground coal mining machinery based on its operational status as described in any one of claims 1-9, characterized in that, include: Anchoring framing unit: Absolute anchor point groups are embedded at equal intervals on both sides of the scraper conveyor, and inertial trajectory sets are collected simultaneously to construct a static reference coordinate frame and complete the initial calibration. Fusion attitude solution unit: performs differential fusion of the absolute anchor point set coordinates and the inertial trajectory set according to the acquisition time sequence, and outputs the real-time attitude matrix; Offset identification unit: Compares the real-time attitude matrix with the historical steady-state template, calculates the deformation drift vector, and locates the tilt segment number; The main cause prioritization unit compares the deformation drift vector with the support step distance sequence and the driving power sequence segment by segment to extract key features characterizing the displacement impact amplitude and power fluctuation complexity. After machine learning, it generates an abnormal driving dominance sequence and completes the source priority ranking. Early warning and suggestion unit: Constructs a set of early warning instructions for operating conditions based on the abnormal driving dominance sequence, and pushes targeted intervention suggestions to the field dispatch terminal.

Images