Coal mine personnel safety behavior analysis method

Abstract

The invention belongs to the technical field of intelligent safety supervision, and particularly relates to a coal mine personnel safety behavior analysis method, which synchronously associates a management instruction, personnel behavior data, an equipment state and environment parameters according to timestamps through a management platform, obtains the current position of a personnel in real time, dynamically judges whether the personnel moves correctly or not, and improves the safety of the personnel. When a person approaches each maintenance position, automatically checking whether the maintenance position has omission or not, if the omission is found, generating a rechecking instruction in real time and pausing subsequent operation authorization, if the omission is not found, carrying out consistency checking on the operation sequence, the operation video content and the standard regulation, and if the checking does not reach the standard, carrying out rechecking on the operation video content and the standard regulation; if yes, second-level early warning is triggered immediately, a recheck mechanism is started, subsequent operation authorization is paused by dynamically monitoring operation compliance and combining authority control, personnel are prevented from entering a subsequent maintenance position with an illegal state, and the problem of multi-position continuous violation which can only be traced afterwards in an existing method is thoroughly solved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent safety supervision technology, specifically a method for analyzing the safety behavior of coal mine personnel. Background Technology

[0002] In coal mine production operations, personnel safety management is the core link to ensure smooth production. The underground environment is complex and changeable, full of many unsafe factors such as fire, gas, and coal dust. Unsafe behaviors of personnel are one of the main causes of safety accidents. Therefore, real-time, accurate and effective monitoring and analysis of the safety behaviors of underground workers, especially maintenance personnel, is a key link to improve the inherent safety level of coal mines.

[0003] Existing methods for managing the safety of coal mine personnel largely rely on location tracking and post-incident handling mechanisms. Specifically, after workers have completed all their tasks, their historical activity trajectories and surveillance video records are retrieved to retrospectively identify whether there were any violations such as working in violation of regulations, lingering in dangerous areas, or not following the prescribed routes.

[0004] However, existing technologies still have the following limitations: 1. Existing methods cannot perform real-time monitoring and immediate intervention during the operation process, and can only trace back afterward. This allows personnel to continuously perform violations at multiple work locations. The system cannot implement mandatory safety audits and process interruptions between procedures, making it impossible to prevent and stop safety hazards in a timely manner.

[0005] 2. During the maintenance operation, the path control lacks dynamic judgment capabilities, which can easily lead to missed or misjudged path deviations. Furthermore, there are shortcomings in the effective arrival verification of maintenance locations and the operation sequence check. Relying solely on manual records or simple geographic fence markings can easily result in missed inspections, skipping necessary preceding locations, and other violations, leading to a failure to guarantee the quality of the operation. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, this invention provides a method for analyzing the safety behavior of coal mine workers, which can effectively solve the problems mentioned in the prior art.

[0007] The objective of this invention can be achieved through the following technical solution: a method for analyzing the safety behavior of coal mine personnel, comprising: deploying IoT sensors and positioning systems underground to collect equipment status data, environmental parameters and personnel location information in real time, and synchronizing and associating management instructions, personnel behavior data, equipment status and environmental parameters according to timestamps through a management platform.

[0008] Assign personnel to perform maintenance tasks, and determine the built-in maintenance routes, the order of each maintenance location, the maintenance objects, and the standard operating time range for the tasks.

[0009] The system can obtain the current location of personnel in real time and dynamically determine whether they are moving in the correct direction toward the maintenance location. If the path deviates, a level one warning will be issued.

[0010] When personnel approach each maintenance location, the system automatically checks whether any previously completed maintenance locations have been missed. If a missed inspection is detected, a supplementary inspection instruction is generated in real time, and subsequent work authorization is suspended.

[0011] The actual work content, test data, and work duration of the completed maintenance locations are compared with standard parameters in real time, including the consistency verification of operation sequence, work video content and standard procedures.

[0012] If the completed inspection of the maintenance location fails to meet the standards, a level-two warning will be triggered immediately and a re-inspection mechanism will be initiated. Subsequent maintenance tasks can only be carried out after the inspection is verified to be qualified.

[0013] Based on real-time trajectory and work status, the system continuously monitors the compliance of personnel behavior and generates dynamic safety evaluation results at each maintenance location. After a preset time interval between maintenance task assignments, a safety behavior analysis report is generated.

[0014] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention collects personnel location, equipment status and environmental parameters in real time through the underground Internet of Things sensor and positioning system, and synchronizes the personnel behavior data according to the timestamp through the management platform, dynamically monitors the compliance of the operation, and suspends the authorization of subsequent operations in combination with the permission control, preventing personnel from entering the subsequent maintenance position with the violation status, completely solving the problem of multiple continuous violations in multiple locations that can only be traced after the fact in the existing method, and transforming the safety hazards from post-disposal to real-time prevention.

[0015] (2) This invention converts the maintenance route into a continuous point sequence. By calculating the deviation angle between the personnel's real-time movement direction vector and the target position coordinate vector, and combining the real-time distance to dynamically select the angle judgment threshold, it solves the problem of lack of dynamic judgment and easy omission and misjudgment in the existing path control. The maintenance position verification adopts a dual method of geofencing and image recognition to prevent only arriving without working. It also compares the actual and standard maintenance sequence to identify violations of skipping the previous position in real time, making up for the lack of work specifications and improving quality and personnel safety.

[0016] (3) This invention designs differentiated re-inspection processes for different types of anomalies, such as incorrect operation sequence and abnormal video verification, and formulates clear judgment standards to improve the anomaly handling mechanism. In addition, it assigns deduction items for violations based on real-time operation data, generates dynamic safety evaluation results, replaces manual periodic scoring, realizes the automation and real-time nature of safety evaluation, and greatly improves the efficiency of personnel safety behavior management. Attached Figure Description

[0017] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the overall implementation steps of the method of the present invention.

[0019] Figure 2 This is a flowchart illustrating the steps involved in determining whether the path of the present invention has deviated.

[0020] Figure 3 This is a schematic diagram of the analysis process for the maintenance locations that the inventors have completed. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 As shown, the present invention provides a method for analyzing the safety behavior of coal mine personnel, including: deploying Internet of Things sensors and positioning systems underground to collect equipment status data, environmental parameters and personnel location information in real time, and synchronizing and associating management instructions, personnel behavior data, equipment status and environmental parameters according to timestamps through a management platform.

[0023] It should be noted that the IoT sensors specifically include: environmental parameter sensors and equipment status sensors. The environmental parameter sensors include, but are not limited to: gas sensors, carbon monoxide sensors, and methane sensors. All of the above sensors comply with coal mine explosion-proof standards and are deployed in key areas such as mining roadways. The equipment status sensors include, but are not limited to: vibration sensors, temperature sensors, and current and voltage sensors. The positioning system adopts UWB positioning technology, and the specific location of personnel is determined by intrinsically safe devices such as positioning cards worn by personnel.

[0024] The management platform receives data streams from various sensors and positioning systems in real time through the OPCUA standard industrial protocol interface. All data packets are appended with a high-precision timestamp upon access to ensure the time consistency of all subsequent correlation analyses.

[0025] Real-time data collection ensures that management can keep abreast of underground dynamics, and the synchronization of timestamps enables management instructions, personnel behavior, equipment status data, and environmental data to form a corresponding relationship, allowing for rapid traceability.

[0026] Assign personnel to perform maintenance tasks, and determine the built-in maintenance routes, the order of each maintenance location, the maintenance objects, and the standard operating time range for the tasks.

[0027] The system can obtain the current location of personnel in real time and dynamically determine whether they are moving in the correct direction toward the maintenance location. If the path deviates, a level one warning will be issued.

[0028] Reference Figure 2 As shown, in a preferred embodiment of the present invention, the specific method for dynamically determining whether it is moving correctly toward the maintenance position to be reached is as follows: the maintenance route is converted into a continuous point sequence according to the maintenance sequence.

[0029] Set the upcoming maintenance location as the target location, and calculate the coordinate vector from the most recently completed maintenance location to the target location.

[0030] Based on the sequence of personnel location points obtained by the positioning system, the current movement direction vector of the personnel is obtained through differential calculation.

[0031] Calculate the deviation angle between the real-time movement direction vector and the coordinate vector.

[0032] The angle determination threshold is dynamically selected based on the real-time distance between the current location of the person and the target location.

[0033] If the deviation angle is greater than the corresponding angle judgment threshold and the deviation duration reaches the preset deviation duration, it is judged as path deviation and a level 1 warning is triggered.

[0034] It should be noted that the built-in maintenance route of the task is broken down into a continuous sequence of coordinate points, with each maintenance location corresponding to a unique coordinate. These points are sorted in the maintenance order to form a standard point sequence. The positioning system collects the coordinates of the intrinsically safe positioning cards worn by personnel in real time at a preset frequency to generate a personnel location point sequence.

[0035] Once personnel have completed the current maintenance position, the next maintenance position is automatically set as the target position. The coordinates of the current maintenance position and the target position are extracted, and the target coordinate vector is obtained through vector operations.

[0036] Based on 3-5 consecutive personnel location points obtained by the positioning system, the differential calculation method is used to first calculate the displacement vector of two adjacent points, and then take the average value of the displacement vector to obtain the current movement direction vector of the personnel. This method can effectively filter out high-frequency positioning jitter and more stably reflect the overall movement trend of personnel.

[0037] The deviation angle between the movement direction vector and the coordinate vector is calculated using the vector dot product formula. After obtaining the current coordinates of the personnel and the coordinates of the target position, the management platform automatically calculates the real-time distance between the two points using the distance formula between two points in three-dimensional space. The vector dot product formula and the distance formula between two points are existing technologies and will not be elaborated here.

[0038] The first, second, and third angle determination thresholds are set in ascending order. When the real-time distance is greater than the preset upper limit distance value, it is marked as a long distance, and the angle determination threshold is set to the third angle determination threshold. When the real-time distance is less than the preset lower limit distance value, it is marked as a short distance, and the angle determination threshold is set to the first angle determination threshold. When the real-time distance is less than the preset upper limit distance value and greater than the preset lower limit distance value, it is marked as a medium distance, and the angle determination threshold is set to the second angle determination threshold.

[0039] The preset distance value, the included angle judgment threshold, and the preset deviation time were all obtained through experimental calibration. Specifically, a roadway section consistent with the actual maintenance route environment was selected, covering short-distance, medium-distance, and long-distance scenarios to ensure that the calibration results were adapted to different maintenance location spacings. A positioning system and intrinsically safe positioning card consistent with the actual deployment were used to ensure that the positioning data accuracy was consistent with the actual application scenario. Multiple sets of distance scenario measurements were conducted. Experimenters moved from the current position to the target position, and the probability that the personnel would miss the target position under different directional angle deviations in each distance scenario was recorded. The directional angle corresponding to the highest probability was used as the included angle judgment threshold for the distance scenario.

[0040] The above solution can promptly identify and trigger a Level 1 warning when personnel move in the wrong direction, preventing personnel from entering dangerous areas, missing key maintenance locations, or delaying work due to path deviation, thereby reducing safety risks and efficiency losses caused by incorrect paths from the source.

[0041] When personnel approach each maintenance location, the system automatically checks whether any previously completed maintenance locations have been missed. If a missed inspection is detected, a supplementary inspection instruction is generated in real time, and subsequent work authorization is suspended.

[0042] This invention converts maintenance routes into a continuous sequence of points. By calculating the deviation angle between the personnel's real-time movement direction vector and the target position coordinate vector, and combining this with the real-time distance to dynamically select the angle judgment threshold, it solves the problems of lack of dynamic judgment and easy omission and misjudgment in existing path control. The maintenance position verification adopts a dual approach of geofencing and image recognition to prevent arriving without working. Furthermore, by comparing the actual and standard maintenance sequences, it identifies violations of skipping previous positions in real time, making up for the shortcomings of manual recording or simple fencing, ensuring standardized operation, and improving quality and personnel safety.

[0043] Reference Figure 3 As shown in a preferred embodiment of the present invention, the automatic verification of whether there are any omissions in the completed maintenance locations specifically involves: dividing the maintenance route into a sequence of geofences that must be passed in sequence, with each fence defined by a range of a preset radius centered on the maintenance location coordinates.

[0044] The system receives personnel location data in real time. When a person enters the geofence of any maintenance location, the system automatically triggers the work video recording function on the personnel terminal, and records the geofence number and entry timestamp, marking it as arrived.

[0045] The operation video transmitted in real time from the personnel terminal is analyzed using image recognition technology to extract image features, including equipment status characteristics and detailed features of specific maintenance parts of the equipment.

[0046] The image features are compared with the standard feature library of the maintenance object at the current maintenance location. If the feature matching degree reaches a preset threshold, the maintenance location is determined to be a valid arrival.

[0047] When personnel enter the fence of the current maintenance location, check whether there is a valid arrival record at the previous fence.

[0048] If no valid arrival record is found, subsequent work authorization will be suspended and a re-inspection instruction will be generated.

[0049] It should be noted that convolutional neural networks are used to perform frame analysis on the operation videos transmitted back from the terminal, extracting equipment status features and detailed features of specific maintenance parts of the equipment. The extracted actual image features are then matched with the standard feature library of the current maintenance location and maintenance object stored in the management platform to calculate the degree of matching.

[0050] The matching degree calculation specifically involves preprocessing the extracted actual image by rotating, scaling, cropping, and grayscale equalization to unify the size and grayscale range.

[0051] Non-numerical features are transformed into computable numerical vectors. The difference between the actual feature values ​​and the standard feature values ​​is calculated using the Euclidean distance algorithm and then normalized. The difference between 1 and the normalized result is used as the similarity. Differential weights are set according to the degree of influence of different features on the maintenance qualification, such as whether the safety feature is the core of the maintenance and whether the operation feature is a key step in the maintenance. Core safety features take precedence over key operation features. The final matching result is obtained by linear weighted fusion.

[0052] Taking the valve opening / closing state as an example of equipment status characteristics and the bolt tightening marks as an example of specific maintenance parts of the equipment, the following matching degree calculation process is given: The angular characteristics of the valve opening / closing state are converted into values ​​of 0-360°. If the actual valve angle is 85°, the standard angle is 90°, and the maximum Euclidean distance is set to 90, the actual Euclidean distance is calculated to be 5, and the similarity is 0.94.

[0053] The number of exposed thread turns of the bolt is converted into a pixel height value. The cosine similarity algorithm is used to compare the actual feature with the vector of standard feature, such as the pixel gradient vector of texture and the contour point vector of shape. The cosine value of the angle between the vectors is calculated, with a value range of 0-1. The closer the cosine value is to 1, the higher the similarity.

[0054] Since the valve's on / off status is the core safety feature and the number of exposed threads after bolt tightening is the key operational feature, the weight of the valve's on / off status is set to 0.6, and the weight of the number of exposed threads after bolt tightening is set to 0.4, and the matching degree calculation is carried out.

[0055] If the matching degree between the actual image features and the standard feature library is greater than or equal to the preset threshold, the personnel not only enter the fence but also perform operations on the object to be inspected, meaning that the inspection location is considered to have been effectively reached. If the matching degree is lower than the preset threshold, it is considered to have been ineffectively reached.

[0056] When personnel are detected entering the geofence of the current maintenance location, the system automatically checks whether there is a valid arrival record for the previous geofence. If there is no valid arrival record, the authorization for all subsequent maintenance work by the personnel is suspended and a supplementary inspection instruction is generated. The instruction clearly indicates the location and requirements to be supplemented and is sent to the personnel terminal and management platform. The authorization for subsequent work is restored only after the supplementary inspection is completed and the arrival is determined to be valid.

[0057] The preset radius ranges from 3 to 8 meters. For areas with narrow lanes and dense equipment, the lower limit can be used; for open areas such as main transport lanes, large chambers, or equipment groups, the upper limit can be used.

[0058] The actual work content, test data, and work duration of the completed maintenance locations are compared with standard parameters in real time, including the consistency verification of operation sequence, work video content and standard procedures.

[0059] In a preferred embodiment of the present invention, the operation sequence consistency check specifically involves: obtaining the timestamps of each maintenance location where personnel have effectively arrived, and constructing the actual maintenance sequence according to the order of the timestamps.

[0060] Obtain the order of each maintenance location built into the task to construct a standard maintenance sequence.

[0061] The actual maintenance sequence is compared with the standard maintenance sequence.

[0062] If any maintenance location in the actual maintenance sequence arrives before the preceding mandatory location in its standard maintenance sequence, it is determined that the operation sequence is incorrect.

[0063] If the operation sequence is determined to be incorrect, an error sequence alarm message will be generated.

[0064] It should be noted that the process of comparing the actual maintenance sequence with the standard maintenance sequence is as follows: traverse each maintenance position in the actual maintenance sequence, check the preceding position of that position in the standard maintenance sequence one by one, and determine whether the current position appears earlier than its preceding position. For example, if a maintenance position in the actual maintenance sequence is marked as B and the arrival time is 14:00, but the preceding position of B in the corresponding standard maintenance sequence is A and the arrival time is 14:20, then the position of B in the actual sequence precedes its standard preceding position A, thus determining that the operation order is incorrect.

[0065] If the operation sequence is determined to be incorrect, the management platform immediately generates a structured sequence error alarm message: the operation sequence is disordered, the abnormal position, the corresponding standard preceding position, and the valid arrival timestamp of the abnormal position. The message is simultaneously pushed to the personnel terminal to remind them to immediately pause the current operation and return to the previous position for re-inspection, ensuring that the abnormality can be detected and corrected in a timely manner.

[0066] In a preferred embodiment of the present invention, the consistency check of the operation video content specifically involves: obtaining a sequence of continuous video frames during the maintenance operation.

[0067] The continuous video frame sequence is analyzed frame by frame to extract the actual operation sequence of personnel, the type of tools used, and the characteristics of the objects being inspected.

[0068] a. Perform time alignment and similarity calculation between the extracted actual operation sequence and the standard operation sequence preset for the current maintenance object in the standard operating procedure.

[0069] b. Identify the types of tools used by personnel and verify their conformity with the standard toolset specified in the procedures.

[0070] c. Analyze the state changes of the object under maintenance before and after the operation, and calculate the difference features of the image frames before and after the operation using computer vision algorithms. The difference features include local structural changes and regional texture feature changes.

[0071] The extracted differential features are compared with the predefined expected state change features of the maintenance operation to determine whether the maintenance operation produces a target state transition that matches the expected state change features, and the verification result is output. The expected state change features of the maintenance operation are constructed by feature learning from the standard maintenance operation process.

[0072] If any of a, b, or c above does not meet the requirements, the video is deemed to be abnormal during the preliminary verification.

[0073] It should be noted that the process of extracting the actual operation action sequence of the personnel is as follows: a three-dimensional convolutional neural network is used to extract the posture of the operator in each frame, and the key point coordinates of consecutive frames are connected to form a spatiotemporal motion trajectory to represent a complete operation action. From the actual operation action sequence and the standard action sequence, the core safety actions that cannot be omitted are selected as anchor points. For example, when inspecting a gas sensor, turning on the detection equipment, reading the sensor data, and turning off the detection equipment are used as anchor points to ensure that the anchor point types of the two sequences are completely corresponding. The time interval between adjacent anchor points in the standard action sequence, for example, 2 seconds, is used as a benchmark to adjust the time axis of the actual action sequence. If the interval in the actual sequence is 3 seconds, it is compressed proportionally to map 3 seconds to 2 seconds to ensure that the key actions of the two sequences are aligned in the time dimension. After the time sequence is aligned, the similarity is calculated by setting differentiated weights according to the existence of the action, the order of the action, and the degree of influence of the action details. The similarity calculation is consistent with the above-mentioned calculation logic of the influence of different features on the inspection qualification. The various similarities are added to obtain the total similarity and compared with the preset similarity threshold. If the total similarity is greater than the preset similarity threshold, it is determined to meet the requirements; otherwise, it does not meet the requirements.

[0074] The YOLO detection algorithm is used to identify tool features within video frames. The extracted tool features are compared with a feature table of common coal mine maintenance tools. If the features match completely, the tool type is initially identified. The actual identified tool type is then compared with a standard toolset. Compliance is determined by whether the required tools are complete and whether any prohibited tools are used. If all the tools used are required standard tools, with no missing or redundant non-standard tools, and no prohibited tools are used, the requirements are met. If any required standard tools are missing or prohibited tools are used, the requirements are not met.

[0075] The specific steps for analyzing the state changes of the object before and after the operation are as follows: obtain the image frame before the operation as the reference frame, and obtain the image frame after the operation as the comparison frame. First, the two images are converted to grayscale. The feature points of the reference frame and the comparison frame are extracted and matched using the scale-invariant feature transformation algorithm. The homography matrix is ​​calculated based on the matched feature points. The comparison frame is subjected to perspective transformation so that it is precisely aligned with the reference frame in space, eliminating the interference caused by small changes in shooting angle and distance.

[0076] Using existing accelerated robust feature algorithms, feature points and descriptors are calculated for the reference frame and the registered comparison frame, respectively. Feature point matching is performed by brute force matching to select feature point pairs with high matching degree. The set of displacement vectors of these feature point pairs represents the local structural changes of the image.

[0077] The two registered images are divided into several sub-regions of fixed size. The local binary pattern histogram of each sub-region is calculated. Then, the chi-square distance between the LBPH vectors of the corresponding sub-regions of the reference frame and the comparison frame is calculated. The magnitude of this distance value represents the significance of the texture feature changes in the region.

[0078] The expected state change characteristics of the maintenance operation are constructed by extracting the above-mentioned difference features from multiple sets of image samples before and after standard maintenance operations and then constructing them through feature learning. Specifically, a large number of known and qualified standard operation image pairs are collected, and the difference features of each pair of images are extracted to form a set of feature vectors. The statistical mean and variance of these feature vectors are calculated to form an expected standard difference feature template.

[0079] For the on-site operation image pair to be verified, after extracting its actual difference features, the similarity between the actual difference feature vector and the pre-stored standard difference feature template is calculated. If the similarity is higher than the preset similarity allowable threshold, it is determined that the maintenance operation has produced the expected target state transition and the output meets the requirements; otherwise, the output does not meet the requirements.

[0080] In a preferred embodiment of the present invention, the consistency check of the operation video content further includes a detection data verification process: directly obtaining sensor readings installed on the device through an Internet of Things interface.

[0081] The above data is uniformly converted into a standard numerical format and associated with the current maintenance location, equipment number, and timestamp.

[0082] Call the standard parameter database corresponding to the current maintenance task and equipment to obtain the safety threshold range for each test data.

[0083] The standardized actual detection values ​​are compared with the corresponding safety thresholds in real time.

[0084] If the actual detected value exceeds the safety threshold, the detected data is immediately determined to be abnormal.

[0085] If the preliminary verification of the above video is normal but the detection data is abnormal, it is determined that the video verification is abnormal. If the preliminary verification of the above video is abnormal, regardless of whether the detection data is normal or abnormal, it is determined that the video verification is abnormal.

[0086] It should be noted that the process of converting the above data into a standard numerical format specifically involves converting the visual feature data extracted from the video frames into actual physical quantities through calibration coefficients. For example, the color of the device indicator light is converted into a numerical code, such as green=1 indicating normal operation, yellow=2 indicating warning, and red=3 indicating malfunction.

[0087] For data with units, such as sensor readings, a format is adopted to separate the numerical value and the unit for storage. For example, the original gas concentration of 0.75%VOL is converted into the numerical value 0.75 and the unit %VOL. When storing, it is labeled as follows: data type: gas concentration, standard value: 0.75, unit: %VOL. The text records the normal state value as 1 and the abnormal state value as 0, so as to ensure that the state can be judged by the numerical value later.

[0088] The establishment of the standard parameter database specifically involves: referring to the recommended standards in the field to which the equipment belongs, clarifying the industry-standard safety thresholds, ensuring that the parameters meet compliance requirements, collecting measured data of similar equipment during long-term normal operation, selecting stable data ranges during faulty operation periods as actual operating benchmarks, and organizing historical fault data of similar equipment to reversely determine the dangerous critical values ​​as the range of safety thresholds.

[0089] Based on standardized numerical values ​​and clear safety thresholds, the comparison logic can replace subjective human judgment, reduce human error, and avoid misjudgment or omission.

[0090] In the maintenance logic, video verification can quickly capture whether there are obvious abnormalities in the equipment. If the video has already preliminarily verified an abnormality, it means that the equipment has presented a problem that can be directly identified. At this time, regardless of whether the test data is normal, the more intuitive video abnormality should be given priority as the core judgment basis. If the video preliminarily verifies that it is normal but the test data is abnormal, it means that the equipment may have a hidden fault, such as the indicator light not changing color, but the internal gas concentration is too high. In this case, the video verification did not capture the hidden abnormality. Essentially, the video preliminarily judges that it does not match the actual state. It needs to be corrected to video verification abnormality to avoid misjudgment due to the video missing the hidden risk.

[0091] It covers both explicit and implicit anomalies. Regardless of whether the video is detected first, as long as there is inconsistency between the video and the data or an anomaly has been detected in the video, it is judged as a video verification anomaly. This avoids the implicit risk of the video being normal but the data being abnormal being overlooked, or the explicit risk of the video being abnormal but the data being normal being overlooked, thus achieving full risk coverage.

[0092] In a preferred embodiment of the present invention, the standard procedure consistency check is specifically as follows: when the positioning system detects that a person has entered the geofence of a certain maintenance location, the timing is automatically started, and when the system detects that the person has left the geofence and moved to the next point, the timing ends, and this time period is recorded as the actual working time.

[0093] If the actual working time is lower than the lower limit of the standard working time range, it will be judged as a warning of reduced working time.

[0094] If the actual operation time exceeds the upper limit of the standard operation time range, it will be judged as an operation timeout warning.

[0095] It should be noted that geofencing can accurately define the physical boundaries of a single maintenance location, ensuring that timing starts when personnel enter and ends when they leave, avoiding errors from manual timing, providing technical support for the accuracy of actual operation time, and thus providing a reliable raw data basis for time comparison, ensuring the objectivity and fairness of the judgment results.

[0096] The standard operating time range is determined through experimental calibration. Multiple sets of historical data, all known to be from the same maintenance location and of the same maintenance task type, are collected. The largest and smallest data sets are removed to obtain valid historical data. The arithmetic mean and standard deviation of the valid historical time data are calculated. The lower limit of the standard operating time is the mean minus one standard deviation, rounded down. The upper limit of the standard operating time is the mean plus one standard deviation, rounded up. For example, if the arithmetic mean is 15 minutes and the standard deviation is 2.5 minutes, then 15 - 2.5 = 12.5 minutes. Rounding down, the lower limit of the standard operating time is 12 minutes. 15 + 2.5 = 17.5 minutes. Rounding up, the upper limit of the standard operating time is 18 minutes. The final standard operating time range is 12-18 minutes.

[0097] If the actual working time is lower than the lower limit of the standard working time range, it means that the maintenance personnel may not have fully performed the necessary operations, such as missing key components, which may lead to potential maintenance quality problems. If the actual working time exceeds the upper limit of the standard working time range, it may be due to problems such as the operation process being stuck, the failure to deal with sudden minor equipment failures in a timely manner, or the personnel's lack of familiarity with the operation, which may cause delays in subsequent maintenance tasks. Therefore, it is necessary to use the standard working time range as a benchmark and determine whether there is an abnormality by comparing the actual time.

[0098] When the actual operation time is lower than the lower limit of the standard operation time range, the operation reduction warning can promptly remind managers to check the maintenance records of that location, preventing personnel from simplifying operations in pursuit of efficiency and ensuring that the operation of each maintenance location meets quality requirements. This reduces equipment failures caused by incomplete maintenance from the source. When the actual operation time exceeds the upper limit of the standard operation time range, the operation overtime warning can enable managers to quickly intervene to investigate the cause of the overtime and solve the problem in a timely manner. This prevents the overtime of a single location from affecting the progress of all subsequent maintenance links and ensures that the overall maintenance task proceeds as planned.

[0099] If the completed inspection of the maintenance location fails to meet the standards, a level-two warning will be triggered immediately and a re-inspection mechanism will be initiated. Subsequent maintenance tasks can only be carried out after the inspection is verified to be qualified.

[0100] In a preferred embodiment of the present invention, the triggering condition for determining that the inspection of the completed maintenance location is not up to standard is an incorrect operation sequence or an abnormal video inspection.

[0101] It should be noted that the triggering conditions of incorrect operation sequence or abnormal video verification are based on the verification objective of comprehensively covering process risks and status risks, ensuring that all core scenarios that may lead to non-compliance with safety standards at maintenance locations are included in the judgment, and there are no judgment loopholes.

[0102] In a preferred embodiment of the present invention, the re-inspection mechanism is specifically as follows: based on personnel positioning information, a re-inspection instruction is sent to the other workers closest to the abnormal location that triggered the level two warning. The instruction includes the abnormality type, the target location number, and the risk parameters to be checked. The other workers closest to the abnormal location refers to the worker whose Euclidean distance to the abnormal location is the shortest after obtaining the location of all workers based on the real-time positioning system.

[0103] When the anomaly type is an incorrect operation sequence, the re-inspection personnel need to check on-site the environmental parameters, equipment operation parameters, and data validity parameters that may cause potential risks due to the disordered sequence. If any parameter exceeds the corresponding preset standard threshold range, the re-inspection will fail. If it does not exceed the corresponding preset standard threshold range, the re-inspection will pass.

[0104] When the anomaly type is video verification anomaly, the re-inspection personnel need to perform a second inspection and re-collect the operation video and result data of the maintenance location.

[0105] If all the results of the second inspection meet the standards, the video is reported as abnormal but the re-inspection is qualified. If any item is not qualified, the re-inspection is reported as unqualified.

[0106] It should be noted that the environmental parameters are used to check the concentration and physical state of hazardous substances in the current work area and the affected adjacent areas, such as carbon monoxide concentration, temperature of the maintenance area, and coal dust concentration.

[0107] The equipment operation parameters are key operating indicators of the equipment that may be misoperated, accidentally triggered, or not in a safe isolation state due to disordered sequence. These parameters include power on / off status, upstream gate switch status of the equipment under maintenance, and motor speed.

[0108] The data validity parameters are used to assess the credibility and validity of recorded data due to incorrect ordering. For example, the data collection time must be consistent with the on-site operation time during the re-inspection, and there should be no abnormalities in the deviation of the parameters from the historical operating baseline of the equipment.

[0109] The corresponding preset standard threshold range is set differentially according to the characteristics of the three types of parameters and is calibrated through experimental data. Specifically: collect multiple groups of data from the same type of equipment operating in normal conditions for a long time, perform data cleaning and statistical analysis. For environmental parameters, take the mean ± 2 times the standard deviation of all normal data points as the preset standard threshold range. For example, if the average operating temperature of a certain equipment is 65°C and the standard deviation is 2°C, then the upper limit of the preset standard threshold is 69°C, and the lower limit of the preset standard threshold is 61°C. For equipment operation parameters, take the maximum and minimum values when it operates stably under rated conditions as the upper and lower limits of the preset standard threshold.

[0110] For data validity parameters, statistically calculate the normal fluctuation range of multiple groups of historical data baselines. The maximum fluctuation value is the upper limit of the preset standard threshold, and the minimum fluctuation value is the lower limit of the preset standard threshold.

[0111] The specific content of the secondary inspection is as follows: the reinspection personnel go to the overhauled positions that do not meet the standards to re - collect videos in accordance with the principle of restoring the operation scenario, ensuring that the operation environment during re - shooting is the same as that during the first overhaul to avoid invalid videos caused by environmental differences. Focus on shooting the points with abnormal videos for the first time, and shoot according to the standard sequence of the overhaul operation to avoid missing key links. After the re - shooting is completed, the reinspection personnel need to immediately conduct a preliminary verification of the videos and data, check whether the videos are clear and coherent, whether the key links are completely covered, whether the result data matches the video content, and check whether the re - shot videos are consistent with the synchronized recorded data. If the re - shot videos clearly cover the key links, the video preliminary verification is passed, and the detection data is normal, it is determined that the video is abnormal but the reinspection is qualified. This result indicates that the abnormal video for the first time is due to shooting operation problems rather than the overhaul itself not meeting the requirements, and the overhaul result is finally valid. If there is any problem with any item, it is determined that the reinspection is unqualified. This result indicates that the secondary inspection still does not meet the validity requirements of the overhaul record.

[0112] Through the underground Internet of Things sensors and positioning system, the present invention can collect the personnel positions, equipment status, and environmental parameters in real - time, synchronously associate the personnel behavior data with the management platform according to the time stamp, dynamically monitor the operation compliance, and combine with the permission control to suspend the subsequent operation authorization and prevent personnel from entering the subsequent overhaul positions with non - compliant status, thoroughly solving the problem of multi - location continuous violations that the existing methods can only trace retrospectively, and transforming the safety hazards from post - event disposal to real - time prevention.

[0113] Based on the real - time trajectory and operation status, continuously monitor the compliance of personnel behavior, generate a dynamic safety evaluation result at each overhaul position, and generate a safety behavior analysis report after a preset time interval from the assignment of the overhaul task.

[0114] It should be noted that the preset duration refers to a certain time interval that is calculated from the moment the maintenance task is completed and then continuously accumulated thereafter. When the time interval expires, a safety behavior analysis report for this maintenance operation will be automatically generated.

[0115] In a preferred embodiment of the present invention, the safety evaluation result specifically refers to: retrieving the personnel's violations for each maintenance location, the violations including: actual operation time exceeding the standard operation time range, path deviation, unqualified re-inspection results, and missed inspections, assigning violation deductions based on the behavior, thereby quantifying the personnel's dynamic safety evaluation result for each maintenance location.

[0116] It should be noted that after each maintenance position is completed, the system automatically summarizes the violations at that position, adds up the corresponding violation deductions, and subtracts the final accumulated deductions from the initial baseline score to obtain the personnel's safety score for that maintenance position.

[0117] The deductions for actual operation time exceeding the standard operation time range, path deviation, unqualified re-inspection results, and missed inspections are determined by expert calibration methods. Specifically, the expert team uses a risk matrix as the core assessment tool to quantitatively assess the risk level of each type of violation from two dimensions: the severity of the consequences and the probability of occurrence, and derives the deduction value accordingly.

[0118] The preset duration is calibrated through experiments. Specifically, multiple typical scenarios matching actual maintenance operations are selected, and multiple candidate time intervals are set as the calibration values ​​for the report generation delay duration. In each experimental scenario, after the maintenance task is completed, the report generation process is triggered according to the above-mentioned multiple candidate time intervals. Outliers are removed from each set of experimental data, such as data loss due to network interruption, while valid experimental data is retained. The average completion rate of each candidate time interval is calculated for the valid experimental data, and candidate values ​​with a completion rate ≥ 99% are selected. The average time difference for viewing the above-selected candidate values ​​is calculated, and the candidate value with the smallest time difference is selected as the preset duration.

[0119] This invention designs differentiated re-inspection processes for different anomaly types, such as incorrect operation sequence and abnormal video verification, and establishes clear judgment standards to improve the anomaly handling mechanism. In addition, it assigns deduction items for violations based on real-time operation data and generates dynamic safety evaluation results, replacing manual periodic scoring, realizing the automation and real-time nature of safety evaluation, and greatly improving the efficiency of personnel safety behavior management.

[0120] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A method for analyzing the safety behavior of coal mine workers, characterized in that, include: Collect equipment status data, environmental parameters, and personnel location information, and synchronize and associate management instructions, personnel behavior data, equipment status, and environmental parameters according to timestamps; Dispatch personnel to carry out maintenance tasks, and determine the maintenance routes, the order of each maintenance location, the maintenance objects, and the standard operating time range within the tasks; Obtain the current location of personnel and determine whether they are moving toward the maintenance location they need to go to. If the path deviates, a level one warning is triggered. When personnel approach a maintenance location, the system automatically checks whether there are any missed maintenance locations. If so, a supplementary maintenance instruction is generated and subsequent work authorization is suspended. If not, the system compares the actual work content, inspection data, and work duration of the completed maintenance locations with standard parameters and conducts a consistency check on the operation sequence, work video content, and standard procedures. If the completed inspection of the maintenance location fails to meet the standards, a level-two warning will be triggered immediately and a re-inspection mechanism will be initiated. Subsequent maintenance tasks will be carried out after the verification is qualified. Based on real-time trajectory and work status monitoring, the system monitors personnel behavior compliance and generates dynamic safety evaluation results at the maintenance location. After a preset time interval between maintenance task assignments, a safety behavior analysis report is generated.

2. The method for analyzing the safety behavior of coal mine workers according to claim 1, characterized in that: The specific method for dynamically determining whether it has moved correctly toward the maintenance location it is about to visit is as follows: The maintenance route is converted into a continuous point sequence according to the maintenance order. Set the upcoming maintenance location as the target location, and calculate the coordinate vector from the most recently completed maintenance location to the target location; Based on the sequence of personnel location points obtained by the positioning system, the current movement direction vector of the personnel is obtained through differential calculation; Calculate the deviation angle between the real-time movement direction vector and the coordinate vector; The angle determination threshold is dynamically selected based on the real-time distance between the current location of the personnel and the target location. If the deviation angle is greater than the corresponding angle judgment threshold and the deviation duration reaches the preset deviation duration, it is judged as path deviation and a level 1 warning is triggered.

3. The method for analyzing the safety behavior of coal mine workers according to claim 1, characterized in that: The automatic verification of whether any completed maintenance locations have been missed specifically involves: The maintenance route is divided into a sequence of geofences that must be passed through in sequence. Each fence is defined by a preset radius centered on the coordinates of the maintenance location. It receives personnel location data in real time. When personnel enter the geofence of any maintenance location, the operation video recording function of the personnel terminal is automatically triggered. At the same time, the geofence number and the entry timestamp are recorded and marked as arrived. The operation video transmitted back in real time from the personnel terminal is analyzed using image recognition technology to extract image features, including equipment status characteristics and detailed features of specific maintenance parts of the equipment. The image features are compared with the standard feature library of the maintenance object at the current maintenance location. If the feature matching degree reaches a preset threshold, the maintenance location is determined to be a valid arrival. When personnel enter the current maintenance location fence, verify whether there is a valid arrival record at the previous fence. If no valid arrival record is found, subsequent work authorization will be suspended and a re-inspection instruction will be generated.

4. The method for analyzing the safety behavior of coal mine workers according to claim 1, characterized in that: The operation sequence consistency check specifically includes: Obtain the timestamps of personnel's effective arrival at each maintenance location, and construct the actual maintenance sequence according to the order of the timestamps; Obtain the order of each maintenance location built into the task to construct a standard maintenance sequence; The actual maintenance sequence is compared with the standard maintenance sequence; If any maintenance location in the actual maintenance sequence arrives before the preceding mandatory location in its standard maintenance sequence, it is determined to be an incorrect operation sequence. If the operation sequence is determined to be incorrect, an error sequence alarm message will be generated.

5. The method for analyzing the safety behavior of coal mine workers according to claim 4, characterized in that: The consistency check of the work video content specifically includes: Acquire a continuous sequence of video frames during the maintenance operation; The continuous video frame sequence is analyzed frame by frame to extract the actual operation sequence of personnel, the type of tools used, and the characteristics of the objects being inspected; a. Perform time-sequence alignment and similarity calculation between the extracted actual operation sequence and the standard operation sequence preset for the current maintenance object in the standard operating procedure; b. Identify the types of tools used by personnel and verify their conformity with the standard toolset specified in the procedures; c. Analyze the state changes of the object under maintenance before and after the operation, and calculate the difference features of the image frames before and after the operation using computer vision algorithms. The difference features include local structural changes and regional texture feature changes. The extracted difference features are compared with the predefined expected state change features of the maintenance operation to determine whether the maintenance operation produces a target state transition that matches the expected state change features, and the verification result is output. The expected state change features of the maintenance operation are constructed by feature learning of the standard maintenance operation process. If any of a, b, or c above does not meet the requirements, the video is deemed to be abnormal during the preliminary verification.

6. The method for analyzing the safety behavior of coal mine workers according to claim 5, characterized in that: The consistency check of the work video content also includes a detection data verification process: Obtain sensor readings directly from the device via an IoT interface; The above data is uniformly converted into a standard numerical format and associated with the current maintenance location, equipment number, and timestamp; Call the standard parameter database corresponding to the current maintenance task and equipment to obtain the safety threshold range for each test data item; The standardized actual detection values ​​are compared with the corresponding safety thresholds in real time; If the actual detected value exceeds the safety threshold, the detected data is immediately determined to be abnormal. If the preliminary verification of the above video is normal but the detection data is abnormal, it is determined that the video verification is abnormal. If the preliminary verification of the above video is abnormal, regardless of whether the detection data is normal or abnormal, it is determined that the video verification is abnormal.

7. The method for analyzing the safety behavior of coal mine workers according to claim 3, characterized in that: The standard procedure conformity check specifically includes: When the positioning system detects that a person has entered the geofence of a certain maintenance location, the timer will automatically start and will stop when the person leaves the geofence and moves to the next point. This time period is recorded as the actual working time. If the actual working time is lower than the lower limit of the standard working time range, it will be judged as a warning of reduced working time; If the actual operation time exceeds the upper limit of the standard operation time range, it will be judged as an operation timeout warning.

8. The method for analyzing the safety behavior of coal mine workers according to claim 6, characterized in that: The trigger condition for determining that the inspection of completed maintenance locations fails to meet the standards is an incorrect operation sequence or an abnormal video inspection.

9. The method for analyzing the safety behavior of coal mine workers according to claim 6, characterized in that: The re-inspection mechanism is as follows: Based on personnel location information, a re-inspection instruction is sent to other workers closest to the abnormal location that triggered the Level 2 warning. The instruction includes: the abnormality type, the target location number, and the risk parameters to be checked. When the anomaly type is an incorrect operation sequence, the re-inspection personnel need to check on-site the environmental parameters, equipment operation parameters, and data validity parameters that may cause potential risks due to the disordered sequence. If any of the parameters exceed the corresponding preset standard threshold range, the re-inspection will fail. If they do not exceed the corresponding preset standard threshold range, the re-inspection will pass. When the anomaly type is video verification anomaly, the re-inspection personnel need to perform a second inspection and re-collect the operation video and result data of the maintenance location; If all the results of the second inspection meet the standards, the video is reported as abnormal but the re-inspection is qualified. If any item is not qualified, the re-inspection is reported as unqualified.

10. The method for analyzing the safety behavior of coal mine workers according to claim 1, characterized in that: The specific safety assessment results are as follows: The system retrieves violations by personnel at each maintenance location, including violations such as actual work time exceeding the standard work time range, path deviation, unqualified re-inspection results, and missed inspections. Based on these violations, points are deducted, thereby quantifying the dynamic safety evaluation results of personnel for each maintenance location.

Images