Coal preparation plant personnel illegal behavior monitoring method based on artificial intelligence

Abstract

The invention discloses a coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, relates to the technical field of violation behavior monitoring, and aims at a coal preparation plant with dense high-risk equipment, complex operation environment and strict safety specification. Carrying out real-time identification and hierarchical linkage processing on unsafe behaviors of personnel in the whole production process; a visible light / infrared camera, a millimeter wave radar and a sound sensor are arranged in a key area, and an equipment control system and environment sensors such as gas and dust are butted; an electronic security fence and a regional characteristic parameter library are set in combination with regional modeling, personnel identity, protective wearing and behavior tracks are obtained based on target detection and human body key point / posture analysis, comparison is carried out according to an operation process and an authority specification library, and semantic judgment is carried out on continuous actions and equipment parameter changes by adopting time sequence attention.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of violation monitoring, and in particular to a coal preparation plant personnel violation monitoring method based on artificial intelligence. BACKGROUND

[0002] The coal preparation plant belongs to an industrial scene with high-risk equipment concentration, complex operation environment and strict safety standards. The whole production process involves high-speed or high-energy equipment such as crushers, scraper conveyors, filter presses and sieves. Once personnel violate the operation rules or enter dangerous areas, mechanical injuries and group risk events are likely to occur.

[0003] The existing safety management still mainly relies on manual inspection, fixed video monitoring, operation permission and interlocking management. However, there are still some problems such as more post-fault accountability, less pre-prevention, difficult on-site monitoring, and lagging risk identification. It is difficult to meet the real-time early warning and disposal needs of personnel violation behavior under complex working conditions. In the typical environment of the coal preparation plant (high dust concentration, variable light, and complex equipment background), single visible light video monitoring is easily affected by shielding, backlight, glare and noise interference, resulting in missed judgment and misjudgment, which leads to insufficient stability of online identification of key unsafe behaviors. At the same time, the traditional monitoring method lacks understanding of continuous action-process semantics, and often regards sequences such as approaching equipment, touching buttons and parameter changes as isolated fragments, making it difficult to make comprehensive judgments in combination with compliance processes (such as maintenance requiring shutdown first), thereby making it difficult to identify complex violations such as unauthorized adjustment of equipment parameters without shutdown into the maintenance area. In terms of risk disposal, the existing early warning is triggered by fixed thresholds, lacking multi-dimensional fusion grading based on personnel behavior, equipment state and environmental parameters, and prone to one-size-fits-all alarms, causing early warning fatigue. There is also a lack of close loop between on-site disposal, equipment linkage and result feedback.

[0004] In view of the above practical constraints, the coal preparation plant needs to build a more robust personnel violation behavior monitoring for the following high-risk and high-frequency scenarios: Device-related: not starting and stopping according to double confirmation, unauthorized parameter change, violation of interlocking, etc.; dangerous area crossing: entering the running area without wearing protective equipment, crossing the warning line, and cleaning at close range during operation, etc.; high altitude / limited space: not wearing a safety belt, not detecting toxic and harmful gases, not ventilating, no monitoring or overtime operation, etc.; material transfer and storage: illegal station in raw coal / fine coal storage and yard, over-limit stacking, lack of protection, etc.; electrical and hot work: unauthorized contact with power distribution cabinet, stacking flammable materials, unlicensed hot work and uncleaned residual heat, etc.; daily passage and protection: not wearing a safety helmet / dust mask, crossing the conveyor belt, staying in the forbidden area, etc.

[0005] In summary, how to face the complex working conditions of the coal preparation plant, while ensuring the real-time and accuracy of the monitoring, improve the semantic understanding of continuous behavior and operation process, and realize the risk level linkage disposal and data tracing, is a technical problem to be solved in the field. SUMMARY

[0006] In view of the above existing problems, the present application is proposed.

[0007] The present application provides a coal preparation plant personnel violation behavior monitoring method based on artificial intelligence to solve the problems of traditional monitoring lag, frequent false reports, weak pre-warning and missing linkage disposal.

[0008] To solve the above technical problems, the present application provides the following technical solutions: The present application provides a coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, which comprises, Step S1, data acquisition and space-time alignment: visible light / infrared cameras, millimeter wave radars, sound sensors, and environmental sensors such as gas, dust and material level are deployed in key areas of the coal preparation plant, and running parameters and permission / interlocking states are obtained from the equipment control system, and multi-source data are time-synchronized and space-calibrated; Step S2, region modeling and electronic fence: electronic safety fences and region characteristic parameter libraries are established in high-risk equipment areas, material transfer and storage areas, high-altitude work areas, restricted spaces, electrical and hot work areas, and plant access areas according to the operation layout of the coal preparation plant; Step S3, personnel and equipment identification: personnel / vehicle targets and personal protective equipment wearing states are identified based on video image segmentation and target detection, and continuous action sequences are obtained by using human key point detection and trajectory tracking; Step S4, process and permission comparison preliminary judgment: personnel position, action and equipment state, permission process specification library are compared, and candidate behaviors that may constitute violations are preliminarily judged and marked; Step S5, time sequence semantic judgment: time sequence attention mechanism is used to analyze the semantic of continuous action and equipment parameter change, and output the violation behavior category; Step S6, risk level evaluation: a risk evaluation model is constructed by integrating personnel behavior, equipment state and environmental parameters to dynamically classify candidate violations; Step S7, graded linkage disposal: according to the risk level, reminders, push, on-site sound and light alarm and equipment linkage (including suspension of operation and regional entrance control) are performed respectively, and video and log are recorded to form a closed loop; Step S8, incremental learning update: when new type of violation or misjudgment sample appears, incremental learning update is performed on the violation identification model based on a small number of newly added samples; Step S9, visualization and traceability: personnel distribution, early warning events and equipment state are displayed in real time on a three-dimensional visualization supervision platform, and statistical reports are generated for safety management and traceability.

[0009] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein the multi-modal perception comprises: The high-definition infrared camera is used to obtain the human body temperature profile in the dust and variable light environment; The millimeter wave radar is used to penetrate the dust to identify the limb action; The sound sensor is used to capture the abnormal sound of the violation operation and to perform triple data verification with the video features to reduce missed judgment and misjudgment.

[0010] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein the region modeling comprises dynamic scene adaptive parameter adjustment: In the large space region, the radar detection range and the camera zoom are automatically lifted, the noise filtering is enhanced in the closed noise area, and the electronic safety fence is set around the equipment to realize the out-of-bound detection and linkage emergency stop / interception.

[0011] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein the identification of personal protective equipment comprises detection of safety helmets, dust masks and hearing protection, wherein the safety helmet is determined by comparing the head profile and the reflective strip and other features, and when not worn according to the specification, the personnel information is recorded and included in the safety assessment at the entrance of the production area.

[0012] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein the violation behavior classification adopts fine-grained feature modeling, divides the personnel violation behavior into multiple categories and multiple sub-classes, and establishes a special feature library and judgment rule for the scene of not wearing a safety helmet / crossing a belt conveyor / adjusting device parameters without authorization / entering the running area / without a safety belt in high altitude / without monitoring in restricted space / detection failure / without ventilation, etc.

[0013] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein the time sequence semantic judgment is performed by modeling the continuous actions of personnel approaching equipment / touching start-stop / equipment parameter change and the relationship with the process flow to determine the composite violation of unauthorized adjustment of equipment parameters in the running state entering the maintenance area, and the safety distance constraint in the industry regulations is used as a hard constraint condition for judgment.

[0014] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein the risk level assessment is based on the multi-dimensional fusion of personnel behavior, equipment state and environmental parameters: When the dust concentration exceeds the standard and it is detected that the dust mask is not worn and the equipment is in the running state, the risk is raised to high risk; The absence of basic protection in the non-operation area is only as a low-risk reminder.

[0015] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein: the hierarchical linkage treatment includes: The low-risk passes through the field voice broadcast and the mobile terminal reminder; The medium-risk is pushed to the area person in charge and triggers the sound and light alarm; The high-risk is reported to the safety management department and is linked with the device control system to suspend operation or close the related entrance, and automatically retains the process video.

[0016] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein: further includes industry knowledge embedding, by converting the coal preparation plant safety regulations and the device safety operation range into knowledge graph and model constraint: When it is detected that the safety distance between the personnel and the running belt conveyor is less than the preset threshold or the filter press operation parameter exceeds the safety range, it is directly determined as a violation and a warning is triggered.

[0017] As a preferred scheme of the coal preparation plant personnel violation behavior monitoring method based on artificial intelligence, wherein: the method runs on the edge computing device or the intelligent camera with the pruned and quantized lightweight model, realizes ≤0.5 second level local real-time identification, only uploads the high-risk events and video clips to the cloud, and labels the risk level with different colors on the three-dimensional visual supervision platform, supports playback and generates periodic statistical reports.

[0018] The present application has the advantages that: the present application combines video and device control system data, compares the authority / process specification library, can reliably identify single person start / stop, unauthorized participation, etc., reduces false positives and missed detections; electronic safety fences are set around key devices, combined with human key point detection, to realize immediate alarm for non-wearing protection crossing and close-range cleaning in operation; a time sequence attention mechanism is introduced to perform semantic analysis on continuous actions such as approaching the device, touching the button, and parameter change, which can identify unauthorized adjustment of device parameters without shutdown into the maintenance area and other complex violations. The risk is predicted by combining personnel behavior, device state and environmental parameters, and the risk is classified as low, medium and high, and the linkage is realized, to realize the closed-loop intervention of reminding, sound and light alarm, linkage shutdown / access control, and significantly reduce false positives and treatment vacuum.

[0019] The application converts the rigid rules such as distance / process of the safety regulations of the coal preparation plant into model constraints and knowledge graphs, directly determines, for example, that the distance between the personnel and the belt during operation is greater than or equal to 1.5 meters, and improves the discrimination accuracy of industry compliance / violation. Through the lightweight model of pruning and quantization, the local recognition is realized in less than 0.5 seconds on the edge device / intelligent camera, only the high-risk segments are uploaded, the real-time performance is ensured, and the dependence on bandwidth is reduced. A three-dimensional supervision platform is constructed, the personnel distribution, early warning and equipment state are displayed in real time, and a periodic statistical report is generated, supporting fine safety management.

[0020] Based on the exclusive data set and incremental learning, the application can update the model with a small number of samples when a new type of violation occurs, so that the system evolves with the scene and reduces the full retraining cost. In the scraper conveyor field, the system triggers an emergency stop in real time without stopping for cleaning, avoiding the accident of being rolled up, and proving that the scheme can realize effective intervention under real working conditions. High-risk events can be synchronized to the safety management department and linked with the equipment control system (pause operation / close entrance), realizing the whole-process closed loop of early warning-intervention-recording. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as limiting the scope of the application.

[0022] Figure 1 The flowchart of the artificial intelligence-based personnel violation behavior monitoring method of the coal preparation plant in the embodiments. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the application more clear and explicit, the following will further describe the application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the application.

[0024] All terms used in the application (including technical and scientific terms) have meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted to have meanings consistent with the context of the specification, and should not be interpreted in an idealized or overly formal manner.

[0025] For example, the terms "first", "second", etc. used in the application are only used to distinguish similar objects, to distinguish the first object from another object, and are not used to describe a specific order or sequence, nor can be understood as indicating or implying relative importance.

[0026] Embodiment 1, reference Figure 1The embodiment provides device operation and dangerous area boundary crossing comprehensive monitoring. There are a large number of high-speed running, high-pressure and strong electromagnetic devices (such as crushers, scraper conveyors, filter presses and screens) in the coal preparation plant, and the core safety hidden danger is that personnel violate the operation or mistakenly enter the dangerous area. The embodiment provides a corresponding monitoring method: Device illegal start and stop and parameter tampering: starting and stopping key devices (such as crushers) without double confirmation process, unauthorized personnel modifying device operating parameters (such as adjusting the speed of the conveyor), and illegally releasing device safety interlocking devices (such as shielding the emergency stop button).

[0027] Monitoring method: capture personnel operation actions (such as single person touching start and stop button) through camera, combine with device control system data (such as parameter abnormal modification record), real-time compare operation permission process specification library, once find unauthorized operation or process missing, immediately trigger sound and light alarm and synchronize to the control room. Avoiding the device overload, jamming caused by misoperation, or the accident of personnel entering the maintenance area during device running caused by interlocking failure.

[0028] Dangerous area illegal intrusion: personnel without wearing protective equipment (such as safety helmet, anti-throw shoes) into device running area (such as inside the guardrail on both sides of the scraper conveyor), crossing device safety warning line (such as the edge of the filter press operation platform), and cleaning up the debris at close range when the device is running.

[0029] Monitoring method: using video image segmentation and human key point detection technology, draw electronic safety fence around the device, real-time identify whether personnel cross the boundary; at the same time, check whether personnel wear protective equipment according to the specification through target detection algorithm, double determine illegal behavior. A coal preparation plant monitored that personnel did not stop the scraper conveyor and stretched out their hands to clean the coal on the conveyor belt when the scraper conveyor was running, immediately triggered the emergency stop signal, avoiding the hand being involved in the accident.

[0030] Embodiment 2 provides high-altitude and restricted space operation monitoring, including: The coal storage warehouse top, thickening pool maintenance, chute cleaning and other operations in the coal preparation plant belong to high-altitude or restricted space operation, which need to strictly follow the process of detecting first, ventilating second and operating third, and comprehensive monitoring can solve the problem of limited visual field and easy fatigue of traditional manual monitoring: High-altitude operation monitoring: personnel do not wear safety belts / rope when working on the top of the coal storage warehouse and the platform of the trestle, cross the gap between the platforms illegally, put debris on the high-altitude operation surface (easy to fall and injure the personnel below), and non-operating personnel stay below the high-altitude operation.

[0031] Monitoring method: Through the deployment of wide-angle cameras around the high-altitude operation surface, combined with human body posture analysis algorithm, identify whether the personnel exist the behaviors of not wearing safety belt (no rope body feature at waist), body leaning out of the platform edge (key points exceeding the boundary of the operation surface), etc.; At the same time, monitor whether there are personnel staying in the area below in violation of the rules, forming double protection of high-altitude ground.

[0032] Restricted space operation monitoring: Before entering the restricted space such as concentration pool and coal slime barrel, toxic and harmful gases (such as hydrogen sulfide) are not detected, ventilation is not carried out, there is no external monitor during operation, and operation time is exceeded without rotation.

[0033] Monitoring method: Video and gas sensor data fusion mode is adopted: on the one hand, through the camera to confirm whether there is a monitor at the entrance of the restricted space and whether the operation personnel carry portable gas detector; on the other hand, the gas sensor data is connected, if the concentration of harmful gas is detected to be over standard or the operation time is exceeded, the comprehensive monitoring immediately triggers the alarm and pushes to the safety management terminal.

[0034] Embodiment 3 provides joint monitoring in the scenes of material transfer and storage, electrical and fire safety, and daily access and protection, such as raw coal warehouse, clean coal warehouse and coal slime yard in coal preparation plant, which are easy to cause accidents such as collapse and dust leakage due to illegal stacking, over-limit storage or inadequate protection, and the comprehensive monitoring can cover the personnel behavior and environmental related risks in such scenes: Illegal operation monitoring of coal storage warehouse: personnel standing illegally on the top of raw coal warehouse (easy to cause the roof cover to fall due to loosening), not controlling the amount of feeding according to the layered stacking specification (leading to the collapse of the coal warehouse due to unbalanced load), and not taking forced ventilation and safety rope fixing measures when entering the coal warehouse to clean the blocked coal.

[0035] Monitoring method: Through the thermal imaging camera (adapted to the environment with much dust and poor light) deployed on the top of the warehouse, whether the personnel stay in the non-operation area on the top of the warehouse is identified; combined with the data of the coal warehouse level sensor, if the level exceeds the safety threshold and is accompanied by personnel approaching illegally, the comprehensive monitoring automatically associates the risk of over-limit personnel, reminds the central control room to adjust the feeding in time and persuade the personnel to leave; for the operation in the warehouse, the safety rope wearing and the ventilation equipment running state are confirmed through the built-in camera.

[0036] Illegal behavior monitoring of coal slime yard: personnel walking illegally on the edge of coal slime pile (easy to be buried by landslide), not spraying dust suppressant according to the regulation (leading to dust exceeding the standard), and non-operation vehicles entering the yard (easy to collide with personnel or crush the pile).

[0037] Monitoring method: Video target classification and regional intrusion detection technology are adopted to distinguish three types of targets, i.e. personnel, vehicles and piles, if it is detected that the personnel enter within 3 meters of the edge of the pile, the vehicle enters without authorization or the personnel are still working when the dust suppression equipment is not started, the comprehensive monitoring immediately generates illegal early warning, which is synchronously pushed to the on-site sound and light alarm and the safety management personnel APP.

[0038] Electric area violation monitoring: non-electrician personnel open the power distribution cabinet door (easy to touch the electricity), personnel put flammable materials (such as cotton yarn, lubricating oil barrels) around the electrical equipment, wet hands touch electrical switches or violate the plug when the equipment is running.

[0039] Monitoring method: through the deployment of infrared thermal imaging and visible light dual-camera cameras in the electrical room, on the one hand, identify whether personnel have unauthorized contact with power distribution cabinets and store flammable materials, and on the other hand, detect whether the electrical equipment is overheating (predict short circuit risk in advance) through infrared thermal imaging. If personnel violate the rules and the equipment is overheated at the same time, the comprehensive monitoring system triggers a high-level alarm.

[0040] Fire operation violation monitoring: without handling the fire operation permit, carrying out welding and cutting operations, no fire extinguisher and fire monitoring personnel around the fire point, and no cleaning of residual fire after the fire (easy to ignite coal dust or flammable materials).

[0041] Monitoring method: through the flame recognition and certificate verification linkage mode: first, identify the flame characteristics generated by the fire operation through the camera, and interface with the enterprise fire operation permission system to check whether there is a valid fire operation permit in the current area; second, confirm whether there is a fire extinguisher within 5 meters of the fire point and whether the fire monitoring personnel are on duty; after the fire operation is completed, the comprehensive monitoring system continues to monitor for 15 minutes (residual fire hazard period), and if residual sparks or personnel are found to have not cleaned up the site, prompt rectification immediately.

[0042] Full protection equipment wearing monitoring: personnel entering the production area without wearing safety helmets, wearing dust masks in dust areas, and wearing earplugs in noise areas (such as fan rooms).

[0043] Monitoring method: through the cameras deployed at the entrance of the factory and the entrance of the workshop, using face and head / face target detection algorithms, automatically identify whether personnel are wearing corresponding protective equipment (such as the hemispherical profile and reflective strip features of safety helmets, and the face blocking features of dust masks), and those who do not wear them are immediately intercepted and reminded, while recording the information of the personnel who violate the rules for safety assessment.

[0044] Factory area violation monitoring: personnel cross the conveyor belt (not use the special channel), stay on the main road of the factory area (affect vehicle traffic), and climb over the safety fence to enter the production area.

[0045] Monitoring method: through the factory video monitoring network, demarcate the electronic fence in the special channel forbidden area, combined with human motion trajectory analysis, if it is identified that personnel deviate from the special channel, enter the forbidden area or climb over the fence, the comprehensive monitoring system pushes the warning to the on-site management personnel in real time, and reminds personnel to correct their behavior through the factory area broadcast.

[0046] In summary, the comprehensive monitoring of the illegal behavior of the personnel in the coal preparation plant is essentially to solve the pain points of the traditional safety management, such as more post-fault accountability, less pre-prevention, difficult on-site monitoring, and lagging risk identification, by means of technical substitution of manual work, the scene coverage follows the risk priority principle, and the links with high casualty risk, high frequency of illegal behavior, and blind area of manual monitoring are focused on, so as to realize the safety management closed loop of real-time identification, automatic alarm linkage, disposal data tracing, and reduce the accident rate in the coal preparation plant.

[0047] In combination with the above embodiments, it can be seen that the present application brings the following significant improvements: (I) Multi-modal perception fusion: breaking through the monitoring bottleneck of complex operation environment 1. Cross-device data collaborative collection Innovative integration of high-definition infrared camera, millimeter wave radar and sound sensor multi-device data: In view of the problems of high dust concentration, variable light (such as humid and reflective in the coal washing workshop, backlight in the belt conveyor area) in the coal preparation plant, the infrared camera captures the personnel temperature trajectory (distinguishes between human body and high-temperature components of equipment), the millimeter wave radar penetrates the dust to identify personnel body movements (avoids misjudgment caused by visual obstruction), and the sound sensor assists in monitoring the abnormal sound of illegal operation (such as abnormal cough when not wearing protective equipment as required, collision sound when operating equipment illegally), forming a triple data verification of vision, radar and acoustics, solving the problems of missed judgment and misjudgment of single visual monitoring in complex environment.

[0048] 2. Dynamic scene self-adaptive adjustment Based on the environmental differences of different operation areas (such as raw coal warehouse, flotation workshop, filter press room) in the coal preparation plant, a dynamic parameter adjustment mechanism of comprehensive monitoring model is constructed: for example, in large space areas such as raw coal warehouse, the radar detection range and camera zoom ratio are automatically increased; in closed areas such as flotation workshop, the noise filtering algorithm of sound sensor is enhanced (filtering machine roar, focusing on personnel conversation and illegal operation sound), realizing intelligent matching of regional characteristics-monitoring parameters, and avoiding the lack of adaptability caused by uniform monitoring standard.

[0049] (II) Illegal behavior identification: from single judgment to semantic understanding 1. Fine-grained illegal behavior classification and feature modeling Break the traditional binary judgment mode of whether it is illegal, and decompose the personnel's illegal behavior in the coal preparation plant into 8 categories and 23 small items (such as protective equipment violation including not wearing a safety helmet, not wearing an anti-static suit, not wearing a dust mask, etc.; operation process violation including crossing the belt conveyor, adjusting the equipment parameters arbitrarily, entering the equipment operation area illegally, etc.), and build a special feature library for each type of behavior: for example, not wearing a safety helmet through the comparison of head profile and safety helmet features (extracting the arc edge and reflective strip features of the safety helmet), crossing the belt conveyor through the human skeleton key point trajectory (identifying the time and space relationship between the leg crossing action and the belt running direction), to realize the accurate classification and positioning of illegal behavior, rather than just labeling the existence of illegal behavior.

[0050] 2. Time sequence behavior semantic analysis Introduce a time sequence attention mechanism to understand the semantics of continuous actions: for example, the traditional monitoring may judge the personnel's action of approaching the equipment, reaching out to touch the button, and changing the equipment parameters as a single action, while the innovative method identifies the illegal behavior of adjusting the equipment parameters arbitrarily by analyzing the logical relationship of the action sequence; at the same time, combined with the operation process of the coal preparation plant (such as the need to shut down the equipment in advance for maintenance, and wear special tools), the continuous actions of personnel entering the maintenance area without stopping (opening the door, entering, and approaching the equipment) are compared with the compliance process to determine the illegal entry into the dangerous area, solving the problem that instantaneous actions are difficult to associate with illegal scenes.

[0051] (Three), early warning and intervention: from post-tracing to pre-prevention and real-time intervention 1. Dynamic risk level prediction Integrate personnel behavior, equipment state, and environmental parameters to build a risk prediction model: for example, when it is monitored that the personnel is not wearing a dust mask and the dust concentration in the area is over standard (through real-time data from the dust sensor), and the equipment is in operation, the risk level is automatically raised to high risk, and immediate warning is triggered; if only the personnel is not wearing a safety helmet but is in a non-equipment operation area (such as a rest area), it is determined as low risk, and only a reminder notice is sent. Through multi-dimensional data fusion, it avoids the fatigue caused by one-size-fits-all warning, and improves the accuracy of warning.

[0052] 2. Graded response and closed-loop intervention Establish a closed-loop mechanism of comprehensive monitoring and early warning-grading push-on-site disposal-result feedback: Low-risk violation (such as not wearing a dust mask): automatically send a text reminder to the mobile phone APP of the illegal personnel and a voice broadcast on site (such as please wear a dust mask immediately, the dust concentration in the current area is over standard); Medium-risk violation (such as approaching the running belt conveyor): in addition to the personnel reminder, it is simultaneously pushed to the mobile phone of the area supervisor (including the location of the illegal personnel, real-time picture), and triggers the on-site sound and light alarm; High-risk violations (such as crossing the belt conveyor, unauthorized operation of equipment): immediately push to the factory safety management department, at the same time link the equipment control system (such as suspend the equipment operation, close the regional entrance), and automatically record the violation process video to provide basis for subsequent disposal, realize the whole process closed loop of early warning-intervention-recording, avoid the loophole of no one to deal after early warning.

[0053] (Four), model optimization: from static training to incremental learning and industry knowledge fusion 1. Construction of coal preparation plant exclusive data set and incremental learning In view of the insufficient adaptability of general comprehensive monitoring model in industrial scene, the exclusive data set of personnel violation behavior in coal preparation plant is constructed (containing 100,000 and labeled samples, covering violation behaviors under different light, dust and equipment background), and incremental learning mechanism is introduced: when a new type of violation behavior (such as violation operation of new equipment) is monitored, the model parameters are updated only through a small number of new samples (50-100), the model evolves with the scene, and the subsequent maintenance cost is reduced.

[0054] 2. Industry knowledge embedded model training The safety regulations of coal preparation plant (such as the requirement of hanging power card for equipment maintenance above the belt conveyor in the Safety Regulations of Coal Preparation Plant) are converted into the constraint conditions of comprehensive monitoring model: for example, when the belt conveyor is running, the safety distance between personnel and belt is greater than or equal to 1.5 meters as a hard constraint, when the personnel enter the range, the violation judgment is triggered directly, avoiding the misjudgment of general object detection model due to not understanding the industry rules (such as misjudging the compliance maintenance as violation). At the same time, through the knowledge graph, the device parameters (such as the safe operation pressure range of filter press) are associated with personnel operation, when the personnel adjust the parameters beyond the safe range, the violation is immediately identified, and the comprehensive monitoring model understands the industry rules.

[0055] (Five), deployment and application: lightweight and visualization adaptation to industrial scene 1. Edge computing lightweight deployment Considering that the network bandwidth of some areas in coal preparation plant (such as underground raw coal bin, remote filter press room) is limited, the comprehensive monitoring model is lightweight processed (through model pruning and quantization, the model volume is compressed by more than 70%), deployed in edge computing device (such as edge gateway, intelligent camera), realizing local real-time identification and early warning of violation behavior (delay ≤0.5 seconds), only the high-risk violation data and video clips are uploaded to the cloud, reducing the network transmission pressure, avoiding the delay of early warning due to network congestion.

[0056] 2. Three-dimensional visualization supervision platform Combined with the three-dimensional model of the coal preparation plant, a real-time monitoring-data visualization-history tracking integrated platform is constructed: in the platform, the personnel distribution in each area, real-time warning of illegal behavior (marked with different colors according to risk levels), and equipment operation status can be intuitively viewed, and by clicking the illegal warning point, the real-time picture and illegal behavior playback can be viewed; at the same time, daily / weekly illegal behavior statistical reports are automatically generated (such as the proportion of illegal behavior of not wearing safety helmets in raw coal warehouse area is 30%, and the main illegal operation in flotation workshop is unauthorized adjustment of reagent ratio), which provides data support for safety management and solves the problems of scattered monitoring data and difficulty in quantitative analysis.

[0057] Finally, it should be pointed out that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

[0058] In addition, those skilled in the art can understand that although some embodiments herein include certain features rather than others included in other embodiments, the combination of features of different embodiments means that it is within the scope of the present application and forms different embodiments. For example, all the above embodiments can be used in any combination. The information disclosed in the background section is only intended to deepen the understanding of the overall background of the present application, and should not be regarded as acknowledging or implying in any form that the information constitutes prior art known to those skilled in the art.

Claims

1. A method for monitoring violations by personnel in coal preparation plants based on artificial intelligence, characterized in that, include, Step S1, Data Acquisition and Spatiotemporal Alignment: Deploy visible light / infrared cameras, millimeter-wave radar, sound sensors, and gas, dust, and material level environmental sensors in key areas of the coal preparation plant, and obtain operating parameters and permission / interlock status from the equipment control system to perform time synchronization and spatial calibration of multi-source data; Step S2, Area Modeling and Electronic Fence: Based on the coal preparation plant's operational layout, establish electronic safety fences and area characteristic parameter databases in high-risk equipment areas, material transfer and storage areas, high-altitude operation areas, confined spaces, electrical and hot work areas, and plant access areas; Step S3, Personnel and Equipment Recognition: Based on video image segmentation and target detection, identify personnel / vehicle targets and the wearing status of personal protective equipment, and use human key point detection and trajectory tracking to obtain continuous action sequences; Step S4, preliminary judgment of process and permission comparison: compare personnel location, action with equipment status and permission process specification library, and make preliminary judgment and mark candidate behaviors that may constitute violations; Step S5, temporal semantic determination: The temporal attention mechanism is used to perform semantic analysis on continuous actions and changes in equipment parameters, and output the category of violation behavior; Step S6, Risk Level Assessment: Integrate personnel behavior, equipment status, and environmental parameters to construct a risk assessment model and dynamically classify candidate violations; Step S7, Tiered and coordinated response: According to the risk level, reminders, push notifications, on-site audible and visual alarms and equipment linkage are executed respectively, and video and logs are recorded to form a closed loop; Step S8, Incremental learning and update: When new violations or misjudged samples appear, the violation identification model is incrementally learned and updated based on a small number of new samples; Step S9, Visualization and Traceability: The 3D visualization monitoring platform displays personnel distribution, early warning events and equipment status in real time, and generates statistical reports for safety management and traceability.

2. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, Multimodal sensing includes: High-definition infrared cameras are used to capture the outline of human body temperature in dusty and changing lighting environments; Millimeter-wave radar is used to penetrate dust and identify body movements; The sound sensor is used to detect abnormal noises from unauthorized operations and performs triple data verification with video features.

3. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, The region modeling includes dynamic scene adaptive parameter adjustment: The radar detection range and camera zoom are automatically increased in large open areas, noise filtering is enhanced in enclosed noisy areas, and electronic safety fences are set up around the equipment.

4. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, The identification of personal protective equipment includes the detection of safety helmets, dust masks and hearing protection. Safety helmets are judged by comparing the head contour with the reflective strip features. If they are not worn in accordance with the regulations, they will be intercepted in real time at the entrance to the production area and the personnel information will be recorded and included in the safety assessment.

5. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, The classification of violations adopts fine-grained feature modeling, which divides personnel violations into multiple categories and subcategories, and establishes a dedicated feature library and judgment rules for scenarios such as not wearing a safety helmet, crossing a belt conveyor, unauthorized adjustment of equipment parameters, entering the operating area, not wearing a safety belt at height, unsupervised confined space, lack of inspection, and lack of ventilation.

6. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, The temporal semantic determination is achieved by modeling the relationship between the continuous actions of personnel approaching equipment, touching to start or stop, and changes in equipment parameters and the process flow; and the safety distance constraints in industry regulations are used as hard constraints in the determination.

7. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, The risk level assessment is based on a multi-dimensional integration of personnel behavior, equipment status, and environmental parameters: The risk is raised to high risk when dust concentration exceeds the standard and no one is detected wearing a dust mask while the equipment is in operation. The lack of basic protection in non-operational areas serves only as a low-risk warning.

8. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, The tiered and coordinated response includes: For low-risk cases, on-site voice announcements and mobile reminders will be provided. The system simultaneously sends notifications to the area manager for medium-risk areas and triggers audible and visual alarms. High-risk situations are reported to the safety management department and linked to the equipment control system to suspend operation or close relevant access points, while automatically saving the process video.

9. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, It also includes industry knowledge embedding, which transforms coal preparation plant safety regulations and equipment safety operation scope into knowledge graphs and model constraints: When the safe distance between personnel and the operating belt conveyor is less than the preset threshold or the operating parameters of the filter press exceed the safe range, it is directly judged as a violation and an early warning is triggered.

10. The method for monitoring violations by personnel in a coal preparation plant based on artificial intelligence as described in claim 1, characterized in that, This method runs on edge computing devices or smart cameras as a lightweight model with pruned and quantized data, uploading only high-risk events and video clips to the cloud; The risk level is marked with different colors on the 3D visualization monitoring platform, which supports playback and generation of periodic statistical reports.

Images