Intelligent monitoring and diagnosing method for working condition of coal mine solid filling mining key equipment

Abstract

The invention provides a coal mine solid filling mining key equipment working condition intelligent monitoring and diagnosing method. The method comprises the steps that equipment multi-source sensing data are obtained in real time; inputting the data into a high-fidelity digital twinborn model and a lightweight deep learning diagnosis model in parallel, and carrying out collaborative analysis and fusion diagnosis; graded early warning is carried out through a multi-mode man-machine interaction system; when a new abnormal mode is diagnosed and is manually confirmed, online incremental learning is automatically triggered, and continuous self-evolution of the diagnosis capability is realized; and driving a predictive maintenance decision and a self-optimization database based on operation and maintenance full-process data of block chain evidence storage. According to the method, accurate real-time sensing of equipment working conditions, abnormal advanced early warning, operation and maintenance decision look-ahead and system intelligent evolution are achieved, and the safety and efficiency of filling mining are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent coal mining and equipment operation and maintenance technology, specifically to an intelligent monitoring and diagnosis method for the operating conditions of key equipment in solid backfilling mining of coal mines. Background Technology

[0002] Coal is my country's primary energy source, and its green, safe, and efficient mining is a crucial cornerstone for the transformation of the energy system. Solid backfilling mining technology, as a typical green mining method, is a key approach to treating coal-based solid waste, controlling surface subsidence, and releasing coal resources under pressure. Hydraulic backfilling supports and multi-hole bottom-discharge scraper conveyors are the core equipment of this technology, and their reliability, stability, and level of intelligence directly determine backfilling efficiency, mining safety, and working face productivity.

[0003] However, the operation and maintenance of existing solid filling equipment generally face the following technical bottlenecks: (1) Incomplete perception of operating conditions: The perception of equipment operating parameters relies on isolated sensors, lacking panoramic, high-precision, and real-time perception capabilities of the coordinated posture of the robotic arm group, the spatial shape of the conveyor, and the stress state. (2) Delayed and passive anomaly diagnosis: For coupling anomalies such as robotic arm interference and conveyor chain jamming, the existing threshold alarm method cannot achieve early identification, often leading to the amplification of faults. (3) Lack of foresight in operation and maintenance decisions: Maintenance strategies are mostly based on periodic inspections or post-event maintenance, failing to conduct predictive maintenance based on the real-time health status and performance degradation trend of the equipment, resulting in high operation and maintenance costs and low equipment utilization. (4) Fixed level of system intelligence: Existing intelligent systems lack the ability to continuously learn and self-optimize from operational practice, cannot adapt to complex and ever-changing geological conditions and process requirements, and are difficult to upgrade to intelligent systems.

[0004] In recent years, digital twins, artificial intelligence, and the Internet of Things (IoT) have provided new ideas for intelligent operation and maintenance of equipment. However, existing research either focuses on digital modeling of general equipment or on purely data-driven algorithm analysis, lacking a systematic solution that deeply integrates profound physical mechanisms with massive operational data and enables continuous autonomous evolution of diagnostic knowledge, specifically addressing the multi-body coupling, strong nonlinearity, and transient operating conditions characteristics of backfilling mining equipment. Therefore, developing a key technology capable of real-time accurate diagnosis, intelligent forward-looking operation and maintenance, and autonomous continuous evolution is of great significance for promoting the development of solid backfilling mining towards advanced intelligence. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to provide an intelligent monitoring and diagnostic method for the operating conditions of key equipment in coal mine solid backfilling mining, enabling the equipment to be known in its entire life cycle, to identify anomalies, to optimize decision-making, and to learn from the system.

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

[0007] A method for intelligent monitoring and diagnosis of the operating conditions of key equipment in coal mine solid backfilling mining, the key equipment including a backfilling hydraulic support and a multi-hole bottom-discharge scraper conveyor; comprising the following steps:

[0008] S1. Real-time Data Acquisition

[0009] Acquire real-time monitoring data from a multi-source sensor network deployed on critical equipment; the real-time monitoring data should include at least the stroke and pressure of the robotic arm's hydraulic cylinders, the attitude angles of key structural components, and the pitch angle of the perforated bottom-discharge scraper conveyor. Yaw angle Roll angle ;

[0010] S2. Dual-model collaborative diagnosis

[0011] Real-time monitoring data is input in parallel into a high-fidelity digital twin model and a lightweight deep learning diagnostic model to obtain the first and second diagnostic indicators, respectively, and then fused to generate a comprehensive diagnostic result and risk level.

[0012] S3. Intelligent Early Warning and Feedback

[0013] Based on the risk level, a multimodal human-computer interaction system is used for graded early warning and interactive control;

[0014] S4. Diagnostic model self-evolution

[0015] When the comprehensive diagnostic results are manually confirmed as a new abnormal pattern, the online incremental learning process is automatically triggered to collaboratively optimize and update the parameters of the deep learning diagnostic model and the high-fidelity digital twin model using the real-time monitoring data.

[0016] S5. Operations and Predictive Decision Support

[0017] The entire operation and maintenance process data, based on blockchain-based notarization, drives the formation of a predictive maintenance decision-making and self-optimizing operation and maintenance database.

[0018] Preferably, the multi-source sensor network includes displacement sensors, pressure sensors, and tilt sensors; the displacement sensors and pressure sensors are deployed on the cylinders of each robotic arm assembly of the filling hydraulic support; the tilt sensors are deployed at key structural nodes and at both ends of the middle trough of each section of the multi-hole bottom-discharge scraper conveyor; the multi-source sensor network adopts an adaptive sampling strategy, dynamically calculating and updating the dynamic threshold used to trigger sampling frequency switching based on the variance of historical data at different process stages.

[0019] Preferably, in step S1, the real-time monitoring data is synchronously transmitted to the ground cloud big data platform and the underground edge server through the industrial ring network and the mine 5G network;

[0020] The high-fidelity digital twin model is deployed on a ground-based cloud big data platform, integrating the closed-loop vector kinematics mechanism and interference discrimination criterion of the filling hydraulic support, and the traction force calculation mechanism and offset discrimination criterion of the multi-hole bottom-discharge scraper conveyor.

[0021] Preferably, the closed-loop vector kinematics mechanism is used to calculate the real-time pose of key components of the filling hydraulic support by the stroke of the robotic arm cylinder, and to establish abnormal working condition discrimination criteria including the interference criticality of the compaction mechanism.

[0022] The traction force calculation theory is used to calculate the chain traction force based on the real-time position and orientation parameters of the perforated bottom-discharge scraper conveyor, and to establish anomaly discrimination criteria for the horizontal and vertical bending offset distance of the perforated bottom-discharge scraper conveyor.

[0023] Preferably, the deep learning diagnostic model is a lightweight model that integrates convolutional neural networks and long short-term memory networks, and is deployed on a downhole edge server; the input of the deep learning diagnostic model is a standardized and time-aligned multi-sensor data matrix, which is used to extract spatial correlations and temporal dynamic features among multiple parameters.

[0024] Preferably, in step S2, the first diagnostic indicator is used as a priori knowledge feature and fused with the second diagnostic indicator at the feature level or decision level.

[0025] Preferably, in step S4, the online incremental learning process includes a conflict arbitration mechanism. When the similarity between newly input abnormal data and the existing database is lower than a set threshold, the deep learning diagnostic model and the high-fidelity digital twin model update are automatically paused and the expert review process is triggered. Incremental learning can only continue after manual confirmation.

[0026] Preferably, in step S3, the human-machine interaction system, based on the comprehensive diagnostic results and risk level, synchronously releases and interacts with multimodal graded early warning information through explosion-proof industrial touch screens deployed underground, mining intelligent mobile terminals, and panoramic screens in the ground command center;

[0027] The mining intelligent mobile terminal integrates an augmented reality-assisted maintenance module, which can scan equipment identifiers to overlay and display real-time operating parameters, 3D fault diagrams, and maintenance guidance animations in a real scene.

[0028] Preferably, in step S5, the predictive maintenance decision is based on the life decay model of the key components of the key equipment and real-time cumulative load data to construct a remaining useful life prediction model and automatically generate a predictive maintenance plan and spare parts requirement list.

[0029] The execution results data of the predictive maintenance plan are input as feedback signals into the online incremental learning process to optimize the parameters of the life decay model and the remaining useful life prediction model.

[0030] Preferably, in step S5, the operation and maintenance database generates digital fingerprints from key event data of the entire operation and maintenance process, including real-time monitoring data, diagnostic reports, and maintenance records, and stores them in a distributed evidence storage database based on a consortium blockchain to form a trusted digital archive of the entire equipment lifecycle; the participating nodes of the consortium blockchain include at least the mine monitoring center, the equipment manufacturer, and the server of a third-party regulatory agency.

[0031] Digital fingerprints are generated using the SHA-256 hash algorithm. Each notarized transaction must obtain consensus verification from a majority of accounting nodes before it can be recorded on the blockchain.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] By employing a multi-source heterogeneous sensor network and an adaptive sampling strategy, a panoramic and accurate perception of the mechanical posture and stress state of the filling equipment is achieved, fundamentally solving the problem of the one-sidedness of traditional monitoring methods. Building upon this, an innovative dual-channel collaborative diagnostic mechanism integrating a high-fidelity digital twin model and a deep learning diagnostic model is constructed, fully leveraging the dual advantages of the high interpretability of the high-fidelity digital twin model and the sensitivity of the deep learning model to latent features. Based on the diagnostic results, the system achieves efficient early warning and maintenance guidance through multimodal human-computer interaction and AR-assisted technology, and relies on blockchain evidence storage and a lifespan prediction model to drive predictive maintenance decisions, significantly reducing unplanned downtime and maintenance costs. Particularly noteworthy is the system's pioneering online incremental learning function integrating a conflict arbitration mechanism, enabling it to autonomously learn from each confirmed anomaly handling, continuously optimizing the diagnostic model and maintenance strategies, achieving dynamic evolution and sustained improvement in the system's intelligence level. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the present invention.

[0035] Figure 2 This is a schematic diagram of the sensor deployment on the filling hydraulic support and the perforated bottom-discharge scraper conveyor.

[0036] in:

[0037] 1. Front top beam tilt sensor; 2. Rear top beam tilt sensor; 3. Rear slip stroke sensor; 4. Scraper conveyor tilt sensor; 5. Compactor arm displacement sensor; 6. Compactor arm tilt sensor; 7. Compactor arm pressure sensor; 8. Rear support arm displacement sensor; 9. Rear support arm pressure sensor; 10. Rear link tilt sensor; 11. Front link tilt sensor; 12. Front support arm pressure sensor; 13. Front support arm displacement sensor. Detailed Implementation

[0038] The invention will now be further described with reference to the accompanying drawings.

[0039] like Figure 1 As shown, a method for intelligent monitoring and diagnosis of the operating conditions of key equipment in coal mine solid backfilling mining is disclosed. The key equipment includes a backfilling hydraulic support and a multi-hole bottom-discharge scraper conveyor. The method includes the following steps:

[0040] S1. Real-time Data Acquisition: Acquire real-time monitoring data from a multi-source sensor network deployed on critical equipment. The monitoring data should include at least the stroke and pressure of the robotic arm's hydraulic cylinders, the attitude angles of key structural components, and the pitch angle of the multi-hole bottom-discharge scraper conveyor. Yaw angle Roll angle Real-time monitoring data is transmitted synchronously to the ground cloud big data platform and underground edge server via the industrial ring network and the mine 5G network.

[0041] S2. Dual-model collaborative diagnosis: Real-time monitoring data is input in parallel into a high-fidelity digital twin model and a lightweight deep learning diagnostic model to obtain the first and second diagnostic indicators, respectively, and then fused to generate a comprehensive diagnostic result and risk level; the fusion diagnosis uses the first diagnostic indicator as a prior knowledge feature and performs feature-level or decision-level fusion with the second diagnostic indicator to improve the credibility and robustness of the comprehensive diagnostic result.

[0042] S3. Intelligent Early Warning and Feedback: Based on the risk level, graded early warning and interactive control are carried out through a multimodal human-computer interaction system;

[0043] S4. Diagnostic Model Self-Evolution: When the comprehensive diagnostic results are manually confirmed as a new abnormal pattern, an online incremental learning process is automatically triggered to collaboratively optimize and update the parameters of the deep learning diagnostic model and the high-fidelity digital twin model using real-time monitoring data.

[0044] S5. Operation and Maintenance and Predictive Decision Support: Based on blockchain-stored data of the entire operation and maintenance process, a predictive maintenance decision-making and self-optimizing operation and maintenance database is formed.

[0045] Furthermore, the multi-source sensor network includes displacement and pressure sensors deployed on the drive cylinders of each robotic arm assembly (such as the front support, rear support, balance, inclined swing angle, and compaction robotic arm) of the filling hydraulic support; key tilt sensors deployed on the top beam and base; and tilt sensors deployed at both ends of the central trough of each section of the multi-hole bottom-discharge scraper conveyor. Specifically, it includes: front top beam tilt sensor 1, rear top beam tilt sensor 2, rear slip stroke sensor 3, scraper conveyor tilt sensor 4, compaction robotic arm displacement sensor 5, compaction robotic arm tilt sensor 6, compaction robotic arm pressure sensor 7, rear support robotic arm displacement sensor 8, rear support robotic arm pressure sensor 9, rear connecting rod tilt sensor 10, front connecting rod tilt sensor 11, front support robotic arm pressure sensor 12, and front support robotic arm displacement sensor 13.

[0046] The network adopts an adaptive sampling strategy, dynamically adjusting the sampling frequency according to different process stages such as frame relocation, compaction, and unloading, in order to improve data validity and system energy efficiency.

[0047] Furthermore, the high-fidelity digital twin model deployed on a ground-based cloud-based big data platform is the core mechanism for achieving accurate diagnosis in this invention. This high-fidelity digital twin model, deployed on a ground-based cloud-based big data platform, integrates the closed-loop vector kinematics mechanism and interference discrimination criteria of the filling hydraulic support, and the traction force calculation mechanism and offset discrimination criteria of the multi-hole bottom-discharge scraper conveyor. For the filling hydraulic support, it integrates a kinematic model based on the closed-loop vector method, capable of calculating key positions such as the inclination angle of the front / rear bearing beam, support height, and tamping head trajectory in real time from the cylinder stroke. For the multi-hole bottom-discharge scraper conveyor, it integrates a traction force calculation model based on the yaw angle and pitch angle of adjacent middle troughs, capable of simulating and calculating the chain traction force under different bending conditions. It can transform real-time perceived data into primary diagnostic indicators such as "interference risk" and "overload risk."

[0048] The closed-loop vector kinematics mechanism is used to calculate the real-time pose of key components of the filling hydraulic support from the stroke of the robotic arm cylinder, and to establish abnormal working condition discrimination criteria, including the interference criticality of the compaction mechanism; the traction force calculation mechanism is used to calculate the chain traction force based on the real-time pose parameters of the multi-hole bottom discharge scraper conveyor, and to establish abnormal discrimination criteria for the horizontal and vertical bending offset distance of the conveyor.

[0049] Furthermore, the deep learning diagnostic model is a lightweight model that integrates convolutional neural networks and long short-term memory networks. It is deployed on a downhole edge server, and its input is a standardized and time-aligned multi-sensor data matrix, which is used to extract spatial correlations and temporal dynamic features among multiple parameters.

[0050] Deep learning diagnostic models learn complex spatial-temporal correlation patterns directly from multi-sensor time-series data, extract latent fault features that are difficult to describe through human-machine experience, and output secondary diagnostic indicators such as "abnormal pressure and flow characteristics".

[0051] Furthermore, the strategy of fusion diagnosis involves fusing the highly interpretable first diagnostic indicator output by the high-fidelity digital twin model with the highly sensitive second diagnostic indicator output by the deep learning diagnostic model at the feature level or decision level. For example, when the digital twin simulation indicates that "the compaction angle is approaching the critical interference point" (first indicator), and the deep learning model also gives a high confidence level of "abnormal cylinder pressure-displacement relationship" (second indicator), the system will comprehensively determine it as "high probability of interference risk" and upgrade the warning level, thereby significantly improving the accuracy and reliability of the diagnosis.

[0052] Furthermore, the online incremental learning mechanism is the core of this invention's intelligent evolution. The online incremental learning process includes a conflict arbitration mechanism. When the similarity between newly input abnormal data and the existing database is lower than a set threshold, model updates are automatically paused and an expert review process is triggered. Incremental learning can only continue after manual confirmation. The process is as follows: When on-site maintenance personnel handle a system alert and confirm it as a new fault mode, such as a previously unrecorded off-center load pattern caused by roof breakage, all time-series data, diagnostic processes, handling measures, and final effects related to this event are automatically labeled, forming a high-quality training sample. The system uses this sample to fine-tune the deep learning diagnostic model and calibrate relevant parameters in the high-fidelity digital twin model, such as the friction coefficient and stiffness coefficient. This process includes a conflict arbitration module; when the similarity between a newly input sample and the existing database is lower than a set threshold, expert intervention is requested to prevent mislearning.

[0053] Furthermore, this includes intelligent early warning and predictive maintenance based on diagnostic results. The human-machine interface system issues early warnings through multiple channels, including underground explosion-proof touchscreens, mining intelligent mobile terminals (including AR-assisted maintenance functions), and ground command screens, based on risk levels. Simultaneously, predictive maintenance decisions are based on the lifespan decay models of key equipment and components, along with real-time cumulative load data, to construct remaining useful life prediction models. This automatically generates predictive maintenance plans and spare parts requirement lists, driving the maintenance model from reactive repair to proactive prevention.

[0054] The mining intelligent mobile terminal integrates an augmented reality-assisted maintenance module, which can overlay and display real-time operating parameters, 3D fault diagrams, and maintenance guidance animations in a real scene by scanning equipment identification.

[0055] The execution results of the predictive maintenance plan are used as feedback signals to feed into the online incremental learning process to optimize the parameters of the life decay model and the remaining useful life prediction model.

[0056] Furthermore, the method also encompasses trusted storage of operational data. The operational database will process key event data from the entire operational process, including real-time monitoring data, diagnostic reports, and maintenance records, into digital fingerprints after hash processing. This digital fingerprints will then be stored in a distributed storage database based on a consortium blockchain, forming a trusted digital archive of the equipment's entire lifecycle. The participating nodes in the consortium blockchain include at least the mine monitoring center, equipment manufacturer, and third-party regulatory agency servers, creating an immutable and fully traceable digital archive of the equipment, providing a reliable basis for quality traceability, insurance claims, and compliance audits. The digital fingerprints are generated using the SHA-256 hash algorithm, and each storage transaction must obtain consensus verification from a majority of the accounting nodes before it can be recorded on the blockchain.

[0057] Example

[0058] This embodiment takes a 260m ultra-long solid filling working face in a coal mine as the background, and applies the method of the present invention to intelligently monitor and maintain the ZC5160 / 30 / 50 type filling hydraulic support and the SGZ800 / 1400 type multi-hole bottom discharge scraper conveyor.

[0059] System Deployment and Data Awareness:

[0060] like Figure 2 As shown, sensors are installed on each support frame, displacement sensors are installed on each robotic arm cylinder to monitor stroke; pressure sensors are installed in the cylinder pressure chambers; and tilt sensors are installed at the four corners of the top beam and base to monitor the pitch and roll angles of key structural components. Tilt sensors are installed in the middle trough of each section of the multi-hole bottom-discharge scraper conveyor to monitor its pitch angle. With yaw angle All sensor data is collected by an intrinsically safe edge gateway, and the sampling rate is adaptively adjusted according to the process. For example, during the frame relocation process, the sampling frequency of all relevant sensors is automatically increased from the basic 5Hz to 50Hz, and then automatically reduced back to 5Hz after the static support stage. The dynamic threshold used to determine the switching of working conditions is automatically calculated and updated by the gateway through online analysis of the variance of the sliding window of the most recent 10 minutes of data.

[0061] Data transmission:

[0062] Real-time monitoring data is transmitted synchronously to the ground cloud big data platform and the underground edge server via the industrial ring network and the mine 5G network.

[0063] Digital twin collaborative diagnosis:

[0064] (1) Construct and run a high-fidelity digital twin model synchronized with the filling physical equipment in the cloud. The core of this model is the integrated mechanism module: the module has a built-in kinematic equation based on the closed-loop vector method. Input the real-time acquired cylinder stroke data into the module, such as the stroke S3 of the rear support robotic arm, and substitute it into its integrated closed-loop vector equation, see equation (1), to calculate the rear top beam sink angle in real time. If the calculated value exceeds the normal range set based on the top plate conditions, the first diagnostic indicator of "abnormal sinking of the rear top beam" is output.

[0065] (1);

[0066] In the formula, For the displacement of the rear support robotic arm, The mounting distance between the rear linkage and the rear support robotic arm on the base is a constant. The length of the rear link is a constant parameter. This is a constant parameter for the length of the upper connecting rod. For the constant parameter of the rear bearing beam length, The pitch angle of the rear-support robotic arm. Let be the angle between the straight line of distance between the rear connecting rod and the rear support robotic arm on the base and the right horizontal plane. The pitch angle of the rear linkage. Let be the constant angle between the upper connecting rod and the right horizontal plane. The pitch angle is the angle of the front link.

[0067] This module incorporates an interference discrimination criterion based on the dynamic trajectory parameter equations of the tamping robot. Real-time acquired attitude angle data of key structural components, such as the tamping angle of the tamping robot, can be input into this module. Substituting the parameters into the dynamic trajectory equation of the compaction robot arm (see equation (2), the real-time trajectory of the compaction head of the compaction robot arm for filling the hydraulic support can be calculated in real time. Based on the calculation results, the model applies predefined logic, such as judging whether the trajectory of the compaction head intrudes into the safe area of ​​interference with the conveyor (see equations (3) and (4), and automatically outputs the first diagnostic indicator such as "high risk level of interference with the compaction mechanism".

[0068] (2);

[0069] In the formula, To solidify the robotic arm's solidification angle, (°); The blanking gap is in mm; , Each corresponding to the consolidation mechanism according to The position of the trajectory in the horizontal and vertical directions after extension, in mm. The center distance for discharge of the multi-hole bottom discharge conveyor is in mm; The length of the extended mechanism is measured in mm.

[0070] (3);

[0071] Equation (3) is the interference judgment before material drop. Due to the different relative postures of the compaction robot arm and the multi-hole bottom discharge conveyor, the material drop gap is affected. The material is likely to fall above the compaction robot arm. The trajectory position of the compaction robot arm in the horizontal and vertical directions interferes with the filling material, resulting in the compaction head being buried.

[0072] (4);

[0073] Equation (4) represents the interference before the compaction robot arm extends. During the compaction process, the compaction robot arm needs to compact multiple times to achieve the required density. The compaction head is prone to colliding with the multi-hole bottom-discharge conveyor, which can restrict the extension of the compaction robot arm or even damage the equipment. In the equation, For safe material discharge gap, mm. , These are respectively strengthening the organization according to The safe position of the extended horizontal and vertical movement trajectory, in mm.

[0074] Meanwhile, the module incorporates a traction force calculation model. It uses the real-time yaw angle difference values ​​of each section of the multi-hole bottom-discharge scraper conveyor, which are sensed in real time. Substituting into the traction force calculation model, as shown in Equation (5), the real-time traction force distribution of the entire chain can be simulated and calculated. The model compares the simulated traction force with the safety threshold, as shown in Equation (6), and outputs a first diagnostic indicator such as "risk of chain traction force overload at section X".

[0075] (5);

[0076] In the formula, This refers to the traction force at the inlet of the middle channel in section n+1. The coefficient of friction between the scraper chain and the central groove. The traction force, N, when the chain enters the middle slot of section 1; For the first , +1 unit difference in yaw angle in the middle slot, (°); when pitch angle When the pitch angle is >0, take "+"; when the pitch angle is >0, take "+"; When the value is less than 0, a "-" is used. In reality, the difference in scraper chain traction between the upward and downward sections of the multi-hole bottom-discharge scraper conveyor is minimal, and ∆ The range is generally 0° to 3°. When calculating, the difference between the two can be ignored, and C can be calculated as a fixed constant.

[0077] (6);

[0078] In the formula, The safety factor is generally taken as 1.2 to 1.3; This is the safety threshold for the scraper chain.

[0079] This module incorporates an offset discrimination criterion. It uses the real-time yaw angle difference values ​​of each section of the multi-hole bottom-discharge scraper conveyor to detect the offset distance. Substituting into the traction force calculation model, as shown in equation (7), the real-time offset of the curved section can be simulated and calculated. The model compares the simulated offset with the safety threshold, as shown in equations (8) and (9), and outputs a first diagnostic indicator such as "offset exceeds limit".

[0080] (7);

[0081] In the formula, denoted as , where is the length of the central groove, in meters (m); and denoted as 'a' is the width of the central groove, in meters (m). The total length of the S-curve segment is given in meters. The maximum horizontal offset distance of this curved segment, in meters (m). This represents the maximum vertical offset of the curved section. For the first Yaw angle of the middle section of the section; For the first , +1 is the difference in yaw angle between the middle section and the channel.

[0082] (8);

[0083] (9);

[0084] In the formula, This represents the maximum horizontal offset of the curved segment at a given moment. This is the maximum permissible horizontal offset for this curved segment. This represents the maximum vertical offset of the curved segment at a given moment. This is the maximum allowable vertical offset distance for this curved segment.

[0085] (2) A lightweight CNN-LSTM fusion diagnostic model deployed on a downhole edge server receives a standardized, time-aligned multi-sensor data matrix in real time, such as a time-series window containing all cylinder pressure, stroke, and tilt angle data from the past 30 seconds. The CNN layer is responsible for extracting spatial correlation features from multiple parameters at the same time, such as "whether the synergistic relationship between the balancing cylinder pressure and the support cylinder pressure is abnormal," while the LSTM layer is responsible for learning the dynamic evolution pattern of these features over time. The model may identify an "early sign of hydraulic system leakage with a confidence level of 85%" that does not reach the threshold but has an abnormal pattern, and output the corresponding second diagnostic indicator.

[0086] (3) The diagnostic fusion center of the fusion diagnostic model executes the fusion strategy. The center receives the first indicator (high interpretability) from the high-fidelity digital twin model and the second indicator (high sensitivity) from the deep learning model. For example, when the digital twin outputs "medium interference risk" (first indicator) for a certain support area, while the deep learning model outputs "abnormal cylinder pressure-displacement relationship, high confidence" (second indicator), the fusion center uses a decision-level weighted fusion algorithm to improve the comprehensive confidence of the two types of indicators, and finally generates a comprehensive diagnostic result of "high probability mechanical interference risk", and dynamically classifies its risk level into "warning" in the three-level early warning.

[0087] Early warning issuance and operation and maintenance execution:

[0088] When the system generates a Level 3 warning, the warning information is simultaneously pushed to the explosion-proof screens on the supports in that area, the shift leader's mobile app, and the ground-based large screen. Maintenance personnel can scan the faulty equipment using the AR function of the mobile app, which can overlay a 3D indication of the fault location and maintenance animation onto the real-world image. Personnel can then complete the maintenance according to the instructions and confirm the results.

[0089] Predictive maintenance and trusted evidence storage for operations:

[0090] Based on the historical chain traction data recorded during this event, the system invoked the Remaining Useful Life (RUL) prediction model to predict that a certain segment of the chain would reach its lifespan threshold in 15 days, automatically generating a preventative replacement work order. The effect data after the work order execution was used as feedback signals to optimize the RUL model itself. All key data from this entire maintenance process were generated with unique hash values ​​and uploaded to a consortium blockchain-based database comprised of nodes from the mining company, equipment manufacturer, and third-party certification authorities, ensuring immutable notarization and creating a reliable equipment lifecycle archive.

[0091] Model self-evolution:

[0092] One day, a new roof condition was encountered at the working face, causing multiple filling hydraulic supports to exhibit an unrecorded pattern of "non-uniform subsidence of the rear roof beam." The initial model diagnosis was "unknown posture anomaly." After confirmation and processing by on-site engineers through a human-machine interface, the complete data package of this event, from the original perception data and diagnostic process to the manual confirmation labels and handling effects, was automatically labeled as a high-quality training sample.

[0093] The diagnostic model automatically triggers an online incremental learning process. This process uses new samples to perform reverse calibration and optimization of relevant mechanistic parameters in the high-fidelity digital twin model, such as the security factor. The deep learning diagnostic model is fine-tuned, and its network weights are updated so that it can automatically identify such new sinking patterns in the future.

[0094] During the learning process, the conflict arbitration module detected a discrepancy between the pattern of new samples and the existing "normal sinking" database. The module automatically paused updates and generated a review request, which was sent to the expert terminal. The learning process only continued after a senior engineer reviewed and confirmed that this was a new working condition, effectively preventing model degradation due to occasional noise or incorrect annotations.

[0095] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring and diagnosis of the operating conditions of key equipment in coal mine solid backfilling mining, the key equipment including a backfilling hydraulic support and a multi-hole bottom-discharge scraper conveyor, characterized in that, Includes the following steps: S1. Real-time Data Acquisition Acquire real-time monitoring data from a multi-source sensor network deployed on critical equipment; the real-time monitoring data should include at least the stroke and pressure of the robotic arm's hydraulic cylinders, the attitude angles of key structural components, and the pitch angle of the perforated bottom-discharge scraper conveyor. Yaw angle Roll angle ; S2. Dual-model collaborative diagnosis Real-time monitoring data is input in parallel into a high-fidelity digital twin model and a lightweight deep learning diagnostic model to obtain the first and second diagnostic indicators, respectively, and then fused to generate a comprehensive diagnostic result and risk level. S3. Intelligent Early Warning and Feedback Based on the risk level, a multimodal human-computer interaction system is used for graded early warning and interactive control; S4. Diagnostic model self-evolution When the comprehensive diagnostic results are manually confirmed as a new abnormal pattern, the online incremental learning process is automatically triggered to collaboratively optimize and update the parameters of the deep learning diagnostic model and the high-fidelity digital twin model using the real-time monitoring data. S5. Operations and Predictive Decision Support The entire operation and maintenance process data, based on blockchain-based notarization, drives the formation of a predictive maintenance decision-making and self-optimizing operation and maintenance database.

2. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 1, characterized in that, The multi-source sensor network includes displacement sensors, pressure sensors, and tilt sensors; the displacement sensors and pressure sensors are deployed on the cylinders of each robotic arm assembly of the filling hydraulic support; the tilt sensors are deployed at key structural nodes and at both ends of the middle trough of each section of the multi-hole bottom-discharge scraper conveyor; the multi-source sensor network adopts an adaptive sampling strategy, dynamically calculating and updating the dynamic threshold used to trigger sampling frequency switching based on the variance of historical data at different process stages.

3. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 1, characterized in that, In step S1, the real-time monitoring data is synchronously transmitted to the ground cloud big data platform and the underground edge server through the industrial ring network and the mine 5G network; The high-fidelity digital twin model is deployed on a ground-based cloud big data platform, integrating the closed-loop vector kinematics mechanism and interference discrimination criterion of the filling hydraulic support, and the traction force calculation mechanism and offset discrimination criterion of the multi-hole bottom-discharge scraper conveyor.

4. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 3, characterized in that, The closed-loop vector kinematics mechanism is used to calculate the real-time pose of key components of the filling hydraulic support by the stroke of the hydraulic cylinder of the robotic arm assembly, and to establish abnormal working condition discrimination criteria, including the interference criticality of the compaction mechanism. The traction force calculation theory is used to calculate the chain traction force based on the real-time position and orientation parameters of the perforated bottom-discharge scraper conveyor, and to establish anomaly discrimination criteria for the horizontal and vertical bending offset distance of the perforated bottom-discharge scraper conveyor.

5. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 3, characterized in that, The deep learning diagnostic model is a lightweight model that integrates convolutional neural networks and long short-term memory networks, and is deployed on a downhole edge server. The input of the deep learning diagnostic model is a standardized and time-aligned multi-sensor data matrix, which is used to extract spatial correlations and temporal dynamic features between multiple parameters.

6. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 1, characterized in that, In step S2, the first diagnostic indicator is used as a prior knowledge feature and fused with the second diagnostic indicator at the feature level or decision level.

7. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 1, characterized in that, In step S4, the online incremental learning process includes a conflict arbitration mechanism. When the similarity between newly input abnormal data and the existing database is lower than a set threshold, the deep learning diagnostic model and the high-fidelity digital twin model update are automatically paused and the expert review process is triggered. Incremental learning can only continue after manual confirmation.

8. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 1, characterized in that, In step S3, the human-machine interaction system, based on the comprehensive diagnostic results and risk level, synchronously releases and interacts with multimodal graded early warning information through explosion-proof industrial touch screens deployed underground, mining intelligent mobile terminals, and panoramic screens in the ground command center. The mining intelligent mobile terminal integrates an augmented reality-assisted maintenance module, which can scan equipment identifiers to overlay and display real-time operating parameters, 3D fault diagrams, and maintenance guidance animations in a real scene.

9. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 1, characterized in that, In step S5, the predictive maintenance decision is based on the life decay model of the key components of the key equipment and real-time cumulative load data to construct a remaining useful life prediction model and automatically generate a predictive maintenance plan and spare parts requirement list. The execution results data of the predictive maintenance plan are input as feedback signals into the online incremental learning process to optimize the parameters of the life decay model and the remaining useful life prediction model.

10. The intelligent monitoring and diagnosis method for the operating conditions of key equipment in coal mine solid backfilling mining as described in claim 1, characterized in that, In step S5, the operation and maintenance database generates digital fingerprints from key event data of the entire operation and maintenance process, including real-time monitoring data, diagnostic reports, and maintenance records, and stores them in a distributed evidence storage database based on a consortium blockchain to form a trusted digital archive of the entire equipment lifecycle; the participating nodes of the consortium blockchain include at least the mine monitoring center, the equipment manufacturer, and the server of a third-party regulatory agency. Digital fingerprints are generated using the SHA-256 hash algorithm. Each notarized transaction must obtain consensus verification from a majority of accounting nodes before it can be recorded on the blockchain.

Images