Multi-target intelligent monitoring and safety early warning method for underground fully-mechanized face

Abstract

The invention belongs to the technical field of coal mine safety production, discloses a multi-target intelligent monitoring and safety early warning method for an underground fully-mechanized face, and aims to solve the problems that existing underground monitoring is weak in multi-target recognition capability, single in alarm logic, poor in working condition adaptability and lack of a linkage mechanism. According to the method, a multi-level framework of a sensing layer, an edge calculation layer, an application layer and a linkage layer is built, a special recognition engine is built based on a pre-training model through underground scene transfer learning and a self-adaptive fine tuning technology, and a sensing-analysis-early warning-response closed loop is formed by dynamically switching detection areas in combination with coal cutting, frame moving and maintenance working conditions. According to the invention, the intelligence and precision of underground monitoring are improved, the false alarm rate is reduced, and the risk disposal time is shortened.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety production technology, specifically to a multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces. Background Technology

[0002] With the continuous improvement of automation and intelligence in coal mining, the operational efficiency of underground fully mechanized mining faces has significantly increased. However, safety remains a core pain point restricting the industry's development. Traditional underground monitoring mainly relies on manual inspections and fixed camera recordings, which have obvious limitations: on the one hand, manual inspections are inefficient, making it difficult to achieve 24-hour uninterrupted coverage, and are easily overlooked due to the limited energy and experience of personnel, such as large coal piles or unretracted side protection. On the other hand, fixed cameras can only record video and lack real-time analysis and intelligent recognition capabilities, making it impossible to promptly detect violations such as personnel not wearing safety helmets or illegally entering dangerous areas.

[0003] Existing underground monitoring systems suffer from problems such as weak multi-target identification capabilities, simplistic alarm logic, and a lack of linkage mechanisms. Most systems can only detect a single target, making it difficult to simultaneously respond to risks from multiple dimensions, including personnel, equipment, and the environment. Alarm methods are often limited to single prompts, failing to differentiate response strategies based on risk levels. Furthermore, the use of static detection strategies and isolated event response mechanisms makes them unsuitable for various operational conditions such as coal cutting, support relocation, and maintenance, leading to false alarms and missed alarms, thus limiting system usability. Therefore, there is an urgent need to develop a rapidly deployable, highly robust, and scenario-adaptable intelligent safety monitoring method to achieve parallel multi-target identification, tiered early warning, and closed-loop response, thereby improving the inherent safety level of coal mines. Summary of the Invention

[0004] Therefore, the purpose of this invention is to provide a multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces, which aims to solve the problems of existing underground monitoring systems having weak multi-target recognition capabilities, simple alarm logic, lack of linkage mechanisms, and poor adaptability to working conditions in complex environments of fully mechanized mining faces.

[0005] To achieve the aforementioned objectives, the technical solution adopted is as follows: A multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces includes the following steps: S1. Build a multi-level system architecture, which includes a perception layer, an edge computing layer, an application layer, and a linkage layer. Each layer achieves data interaction through a mining ring network, 4G or 5G communication to ensure the stability and timeliness of information transmission. S2. A general computer vision model library is adopted, and a dedicated recognition engine is built through underground scene transfer learning and adaptive fine-tuning technology. Multiple detection areas are configured to achieve parallel recognition of multiple targets. Parallel recognition of multiple targets includes large coal piles in the scraper conveyor area, personnel wearing safety helmet status, personnel intrusion in dangerous work areas, and hydraulic support side protection retraction status. S3. Establish a composite event judgment and alarm classification mechanism, establish a spatiotemporal correlation rule base, perform fusion judgment on multi-source heterogeneous data, define a first-level single event alarm and a second-level combined event alarm, and trigger a preset linkage response for the second-level alarm; S4. Based on the working condition state machine and the regional weight allocation algorithm, the corresponding detection area is dynamically enabled or disabled according to the current working status of the longwall mining face to achieve working condition adaptive detection. S5. After the alarm information is analyzed and processed by the application layer, it is synchronously pushed to the linkage layer to complete the "perception-analysis-early warning-response" closed loop.

[0006] As a further improvement of the present invention, the sensing layer includes an infrared sensor, a hydraulic support status sensor, and a mine explosion-proof camera, used to collect data on the underground environment, equipment status, and personnel behavior.

[0007] As a further improvement of the present invention, the edge computing layer includes an AI edge computing host, which deploys a lightweight inference engine to support adaptive optimization of general models and is used to perform real-time analysis of data collected by the perception layer. The optimization directions include enhancing the reflective features of helmets in low-light environments and improving the robustness of the identification of protective mechanical structures in coal dust interference scenarios.

[0008] As a further improvement of the present invention, the detection area configuration of the multi-target parallel recognition is as follows: the large coal detection area is set at the head and tail of the machine, the personnel intrusion and safety helmet detection area covers the entire working face area, and the side protection detection area is set at two monitoring points in front according to the position of the coal mining machine and the direction of travel.

[0009] As a further improvement of the present invention, the first-level single-event alarm includes alarms for not wearing a safety helmet, personnel intrusion, large coal accumulation, and unretracted guardrails; the second-level combined event alarm is a high-risk alarm of "unretracted guardrails + personnel approaching" triggered by spatiotemporal trajectory prediction and relative speed threshold determination.

[0010] As a further improvement of the present invention, the linkage response of the secondary alarm includes automatically triggering the underground voice broadcast risk warning, pushing the alarm pop-up and video linkage information to the ground dispatch center, and establishing an equipment response priority strategy: 1) voice broadcast alarm; 2) scraper conveyor speed reduction operation; 3) coal mining machine emergency stop + scraper conveyor interlock.

[0011] As a further improvement of the present invention, the fully mechanized mining face operation status includes three modes: coal cutting, support shifting, and maintenance. Based on the state machine driving and resource weight dynamic allocation mechanism, the detection areas activated in different modes are dynamically switched according to preset rules.

[0012] As a further improvement of the present invention, the linkage layer includes an underground voice broadcasting system, a ground dispatch center and an equipment linkage control module, and the equipment linkage control module can realize the interlocking function of the scraper conveyor.

[0013] The beneficial effects of this invention are: 1. Multimodal event coupling recognition: By injecting prior knowledge of downhole features through transfer learning, the recognition degradation caused by dust and low light is solved, and the recognition of multi-dimensional targets such as personnel, equipment and environment is realized simultaneously; 2. Hierarchical response chain design: A spatiotemporal correlation rule base is established to couple and model complex risks. The three-level progressive response chain (voice broadcast → scraper conveyor deceleration → equipment lockout) significantly shortens the handling time of high-risk events. 3. Flexible resource scheduling architecture: The detection area is dynamically adjusted according to different operation modes such as coal cutting, frame moving, and maintenance, avoiding false alarms during non-operation periods caused by fixed detection modes, and improving the system's practicality and reliability.

[0014] 4. Multi-level closed-loop management to enhance inherent safety: Through the collaborative work of the perception layer, edge computing layer, application layer, and linkage layer, a closed-loop process of "data acquisition - real-time analysis - hierarchical early warning - linkage response" is achieved, reducing manual intervention, improving the intelligence level of monitoring and early warning, and fundamentally ensuring the safety of underground fully mechanized mining operations. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a diagram showing the overall architecture of the system described in this invention; Figure 2 This is a flowchart of the alarm logic of the system described in this invention; Figure 3 This is a schematic diagram of the multi-target recognition region configuration of the system described in this invention; Figure 4 This is a preview of the software's main interface. Detailed Implementation

[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0018] A multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces includes the following steps: S1. Build a multi-level system architecture, which includes a perception layer, an edge computing layer, an application layer, and a linkage layer. Each layer achieves data interaction through a mining ring network, 4G or 5G communication to ensure the stability and timeliness of information transmission. The perception layer consists of infrared sensors, hydraulic support status sensors, and mine explosion-proof cameras. It is used to comprehensively collect underground environmental parameters, operating status data of hydraulic supports and other equipment, and behavioral data of workers, providing a basic data source for subsequent analysis.

[0019] Edge computing layer: This includes AI edge computing host and Hikvision AI open platform pre-trained model library. The pre-trained models cover head detection model and object detection model, which can perform local real-time analysis and processing of video and sensor data collected by the perception layer, reducing data transmission latency.

[0020] Application layer: Core functions include multi-target parallel recognition, composite event judgment, and dynamic detection area configuration, realizing the core logical operations of target recognition, risk classification, and working condition adaptation.

[0021] Linkage layer: It consists of an underground voice broadcasting system, a ground dispatch center and an equipment linkage control module. The equipment linkage control module has a scraper conveyor interlocking function and is used to receive alarm commands from the application layer and execute corresponding response actions.

[0022] S2. A general computer vision model library is used for transfer learning and adaptive fine-tuning. Multiple detection areas are configured to achieve parallel recognition of multiple targets. Parallel recognition of multiple targets includes large coal piles in the scraper conveyor area, personnel wearing safety helmets, personnel intrusion into hazardous work areas, and the retraction status of hydraulic support side guards. Targets to be identified include large coal piles in the scraper conveyor area, the status of personnel wearing safety helmets, personnel intrusion into hazardous work areas (between supports, near the coal mining machine), and the status of hydraulic support side protection being retracted.

[0023] Detection area configuration: The large coal detection area is located at the head and tail of the machine to specifically monitor abnormal accumulation in key areas of the scraper conveyor; the personnel intrusion and safety helmet detection area covers the entire working face to achieve comprehensive supervision of the workers; the support detection area is dynamically set at two monitoring points in front according to the position and direction of the coal mining machine to ensure accurate monitoring of the hydraulic support support status.

[0024] S3. By integrating the data from the guardrail status sensor and personnel trajectory prediction through the spatiotemporal association rule base, a first-level single-event alarm and a second-level combined event alarm are defined. When the distance between the personnel and the unretracted guardrail is ≤ the safety threshold and the relative speed is > 0, the high-risk composite event response chain is activated. Level 1 Single Event Alarm: Triggered for a single risk scenario, including alarms for not wearing a safety helmet, personnel entering a dangerous area, large coal piles, and hydraulic support side protection not being retracted, used to alert to routine safety hazards.

[0025] Level 2 Combined Event Alarm: This alarm is triggered in high-risk, complex scenarios, specifically the combination of "failed to retract safety guards + approaching personnel". Such risks are prone to causing major safety accidents and require the activation of the highest level of response.

[0026] Tiered response rules: Level 1 alarms only trigger basic prompt functions; Level 2 alarms trigger preset linkage responses, including automatically starting underground voice broadcasts to provide risk warnings, pushing alarm pop-ups and real-time video linkage information to the ground dispatch center, and generating suggestions to suspend the operation of the coal mining machine to support dispatchers' decision-making.

[0027] S4. Based on state machine-driven and regional weight allocation algorithm, the detection area is dynamically configured according to coal cutting, frame moving and maintenance modes; Operation status classification: including three core modes: coal cutting, frame moving, and maintenance. The risk points and monitoring focus differ in each mode.

[0028] Dynamic configuration logic: Preset the corresponding rules for different operation modes and detection areas. The system can identify the current operation status in real time and automatically switch the detection area. For example, in the coal cutting mode, the detection area for large coal pieces and the side protection detection area is activated. In the maintenance mode, the detection area for personnel intrusion and wearing safety helmets is activated to ensure that the monitoring resources are accurately matched with the working conditions.

[0029] S5. After the alarm information is analyzed and processed by the application layer, it is synchronously pushed to the linkage layer to complete the "perception-analysis-early warning-response" closed loop.

[0030] After the sensing layer collects multi-dimensional data from downhole, it transmits the data in real time to the edge computing layer for localized analysis and processing. The edge computing layer completes target recognition through a pre-trained model and pushes the recognition results to the application layer; The application layer performs composite event judgment on the identification results, triggers corresponding alarms according to the risk level, and dynamically adjusts the detection area based on the current working conditions; Alarm information is simultaneously pushed to the linkage layer, and underground voice broadcasts, ground dispatch centers, and equipment linkage control modules respond synchronously, forming a closed-loop management of the entire process to ensure timely handling of risks.

[0031] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, component splitting or combination, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces, characterized in that, Includes the following steps: S1. Build a multi-level system architecture, which includes a perception layer, an edge computing layer, an application layer, and a linkage layer. Each layer achieves data interaction through a mining ring network, 4G or 5G communication. S2. A general computer vision model library is adopted, and a dedicated recognition engine is built through underground scene transfer learning and adaptive fine-tuning technology. Multiple detection areas are configured to achieve parallel recognition of multiple targets. Parallel recognition of multiple targets includes large coal piles in the scraper conveyor area, personnel wearing safety helmet status, personnel intrusion in dangerous work areas, and hydraulic support side protection retraction status. S3. Establish a composite event judgment and alarm classification mechanism, establish a spatiotemporal correlation rule base, perform fusion judgment on multi-source heterogeneous data, define a first-level single event alarm and a second-level combined event alarm, and trigger a preset linkage response for the second-level alarm; S4. The edge computing layer dynamically loads the corresponding model into memory for different monitoring areas based on the current working status of the longwall mining face, so as to realize adaptive detection of working conditions. S5. After the alarm information is analyzed and processed by the application layer, it is synchronously pushed to the linkage layer to complete the "perception-analysis-early warning-response" closed loop.

2. The multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces according to claim 1, characterized in that: The sensing layer includes infrared sensors, hydraulic support status sensors, and mine explosion-proof cameras, used to collect data on the underground environment, equipment status, and personnel behavior.

3. The multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces according to claim 1, characterized in that: The edge computing layer includes an AI edge computing host, which deploys a lightweight inference engine to support adaptive optimization of general models. It is used to perform real-time analysis of data collected by the perception layer. The optimization directions include enhancing the reflective features of helmets in low-light environments and improving the robustness of the identification of protective mechanical structures in coal dust interference scenarios.

4. The multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces according to claim 1, characterized in that: The detection area configuration for the multi-target parallel identification is as follows: the large coal detection area is set at the head and tail of the machine, the personnel intrusion and safety helmet detection area covers the entire working face, and the side protection detection area is set at two monitoring points in front according to the position and direction of the coal mining machine.

5. The multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces according to claim 1, characterized in that: The first-level single-event alarm includes alarms for not wearing a safety helmet, personnel intrusion, large coal accumulation, and failure to retract the support frame; the second-level combined event alarm is when the front support frame is not retracted while the coal mining machine is in operation, and a personnel target is detected in the same spatial and temporal area, and this event is judged as a high-risk collision event.

6. A multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces according to claim 1 or 5, characterized in that: The linkage response of the level 2 alarm includes automatically triggering underground voice broadcast risk warnings, pushing alarm pop-ups and video linkage information to the ground dispatch center, and generating a suggestion to suspend the operation of the coal mining machine.

7. The multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces according to claim 1, characterized in that: The fully mechanized mining face operation status includes three modes: coal cutting, support shifting, and maintenance. Based on the state machine-driven and resource weight dynamic allocation mechanism, the detection areas activated in different modes are dynamically switched according to preset rules.

8. A multi-target intelligent monitoring and safety early warning method for underground fully mechanized mining faces according to claim 1, characterized in that: The linkage layer includes an underground voice broadcasting system, a ground dispatch center, and an equipment linkage control module. The equipment linkage control module can realize the interlocking function of the scraper conveyor.

Images