A kind of remote supervision-based hot work safety management system and method

Abstract

This invention relates to the field of remote monitoring and discloses a hot work safety management system and method based on remote monitoring. One method includes: constructing a preliminary scenario for the work location, fire source intervention method, and isolation measures; implementing a progressive tracking mechanism for the triggered state of the work access framework to reconstruct the changes in the hot work process; comprehensively analyzing images, sounds, warning zone markings, and personnel disturbances transmitted from the site using a work deviation recognition model; digitally labeling the work units and impact ranges corresponding to protection failure scenarios in the background monitoring interface to form a hot work traceability ledger for file location and reverse tracing of responsibility links; and dynamically adjusting the marking colors, prompts, and inspection priorities in real time based on the dynamic updates of the hot work traceability ledger to generate management instructions for the construction site. This invention has the advantage of improving regulatory transparency.

Description

Technical Field

[0001] This invention relates to the field of remote monitoring, specifically to a hot work safety management system and method based on remote monitoring. Background Technology

[0002] In nighttime renovations of large public shopping malls, operations such as tenant eviction, advertising structure removal, and air conditioning pipe rerouting often require hot work including welding and cutting. Due to the numerous leased areas and frequent overlapping construction activities, management units typically still rely on paper-based hot work permits and telephone reporting for application and on-site confirmation. However, in practice, the lag in updating the status of hot work operations is a typical and prominent problem. Taking a pipe renovation project in a shopping mall as an example, after the construction team submits a paper hot work application, the on-site supervisor informs management personnel by phone that work has begun. However, due to the lack of a real-time information feedback mechanism, supervisors cannot know whether fire barriers are erected as required, whether personnel have entered the restricted area, or whether work conditions have deviated in a timely manner. Shopping malls often have multiple hot work locations simultaneously at night, and the traditional method of relying on manual reporting makes it difficult to create continuous and auditable work records, resulting in management personnel often being unable to quickly locate specific work areas when receiving alarms or inspection reminders. Therefore, it is essential to design a remote-monitoring-based hot work safety management system and method to improve regulatory transparency. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a hot work safety management system and method based on remote monitoring, which has the advantage of improving regulatory transparency and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goal of improving regulatory transparency, this invention provides the following technical solution: a method for safe management of hot work operations based on remote monitoring, comprising the following steps: A preliminary scenario is constructed for the work location, fire source intervention method, and isolation measures. The reporting link is rearranged in a conditional order to generate a work access framework that can be recognized by the scheduling engine. The operation access framework triggers a progressive tracking mechanism, which reconstructs the operation process changes of hot work activities by calling the event remnant set of historical renovation areas and on-site inspection fragments. The operation offset recognition model comprehensively analyzes the images, sounds, warning zone markings and personnel disturbances transmitted from the site. Taking the changes in operation rhythm depicted by the clues of operation process changes as a reference, it identifies the sudden nodes of local safety parameters and automatically infers the protection failure scenarios that occur. The work units and impact range corresponding to the protection failure scenarios are digitally marked in the background monitoring interface. The responsible persons, approval trajectories and execution evidence within the scope are integrated to form a hot work traceability ledger for file location and reverse tracing of responsibility links. Based on the dynamic updates of the hot work traceability log, the identification colors, prompts, and inspection priorities are adjusted in real time to generate management instructions for the construction site.

[0005] Preferably, the process of generating a job admission framework that can be recognized by the scheduling engine is as follows: By analyzing the location of the work site using on-site geographic information, construction layout map and safety isolation points, we can identify areas where combustibles accumulate, confined spaces and ventilation paths. The operational chain of fire source intervention methods is broken down, and fire source characteristic groups are formed based on the intervention point, direction of action and heat transfer characteristics. The consistency of the isolation measures deployment was verified, including the continuity of the isolation zone, the integrity of the fire extinguishing facilities, and the status of locks and tags. Based on the fire source characteristic group, the operation condition nodes are constructed. The approval roles, time sequence and mutually exclusive conditions in the reporting link are rearranged in a partial order. The rearranged condition nodes are then used to generate the operation access framework in a structured manner.

[0006] Preferably, the process of triggering the progressive state tracking mechanism for the job admission framework is as follows: Based on the triggering conditions set in the job admission framework, a status monitoring unit is built for each condition node; The monitoring unit continuously records reporting actions, approval processes, changes in on-site layout, and the on-site status of workers. By using a progressive tracking queue, the trigger time, responsible party, and operation content are uniformly transcribed into state evolution fragments. The state evolution fragments are then sequentially checked to eliminate duplicate actions and fill in missing nodes.

[0007] Preferably, the process of reorganizing the operational process changes of hot work activities is as follows: Retrieve event remnants that match the current work area from the historical modification area database, including equipment modification records, violation fragments, and accident warning items; Align the event remnant set with the on-site inspection footage by time tag, and perform image quality assessment to remove blurry segments; Feature extraction is performed on the protective arrangements, operational actions, and surrounding interference factors that appear in the patrol segments to form a sequence of feature nodes; By combining typical violation patterns in the event residue set, similarity clustering is performed on the feature node sequence to identify key change points in the operation activities and output clues of operation process changes sorted by time.

[0008] Preferably, the process of comprehensively analyzing the images, sounds, warning zone markings, and personnel disturbances transmitted from the site using the operation offset recognition model is as follows: The images transmitted from the scene are preprocessed according to the viewpoint, lighting and local occlusion, the key fire source area and the operation boundary contour are extracted, and the sound signal is decomposed into noise spectrum to identify specific frequency bands associated with abnormal knocking and combustion instability. The integrity and continuity of warning zone markings are detected using a structured recognition model, the area boundary offset is generated, the trajectory of personnel position disturbance is fitted, and the walking path, dwell time and potential boundary crossing behavior are analyzed. Images, sounds, warning zone markers, and personnel disturbances are fused together with timestamps and spatial coordinates to form a multimodal input set for offset detection.

[0009] The preferred process for automatically inferring potential protection failure scenarios is as follows: Align the temporal features in the multimodal input set with the cues of changes in the work process, and calculate the work rhythm offset at different stages; Dynamic thresholds are established for the speed of heat source approach, the range of personnel movement, and the degree of warning zone deviation. A sudden jump detection method is used to identify discontinuities and rapid spikes in parameter changes. The system categorizes leap nodes by risk type and automatically infers potential protection failure scenarios based on the spatial distribution of leap nodes and changing links.

[0010] Preferably, the process of digitally labeling the work unit and its impact range corresponding to the protection failure scenario in the background monitoring interface is as follows: Map the scenario types in the protection failure scenario to the corresponding work units, including fire source unit, isolation unit and patrol unit; The scope of risks involved in the scenario is expanded using a spatial grid to form a visualized area of ​​impact; The work units and affected areas are digitally labeled in the regulatory interface, and the labeling content includes risk level, trigger time and relevant fragment index.

[0011] Preferably, the process of creating a hot work operation traceability ledger for archival location and reverse tracing of responsibility is as follows: The information of the responsible parties in the digital annotations is matched with the roles and process nodes in the approval system; The action records, approval timestamps, and execution evidence of the responsible parties are structured and integrated to form responsibility chain segments. All responsibility chain segments are then strung together in chronological order to fill in any missing approval or inspection nodes. By introducing reverse backtracking rules, the responsibility chain is traced backward from the latest situation to locate the link of responsibility and form a hot work operation traceability ledger for file location and reverse backtracking of the responsibility chain.

[0012] Preferably, the process of generating management instructions for the construction site is as follows: Based on the risk evolution status in the traceability ledger, the risk level of each marked element in the current regulatory interface is dynamically refreshed. Different color and prompting methods are applied to the labeled elements of different risk levels, including flashing, voice prompts or vibration prompts; The patrol routes are reordered, and high-risk areas are highlighted in advance along the patrol path, generating management instructions for the construction site.

[0013] This invention also discloses another technical solution: a hot work safety management system based on remote monitoring, comprising: Work access module: The location elements, ignition source methods and isolation and protection arrangements for hot work are structured to generate a work access framework; Status tracking module: After the access framework is triggered, retrieve historical event remnants and inspection records to reconstruct clues of changes in the operation process; Risk identification module: Integrates images, sounds, warning zone markings, and personnel disturbances to infer protection failure scenarios based on clues of changes in the work process; Traceability Ledger Module: Digitally marks protection failure scenarios, integrates responsible persons, approval trajectories and execution evidence to form a hot work operation traceability ledger; Command scheduling module: dynamically adjusts and generates management commands based on the updated results of the hot work traceability log.

[0014] Compared with the prior art, the present invention provides a hot work safety management system and method based on remote monitoring, which has the following beneficial effects: This invention achieves real-time modeling and dynamic verification of the entire hot work process by constructing a work access framework recognizable by the scheduling engine, introducing a progressive tracking mechanism based on trigger states, and combining historical event remnants with on-site inspection fragments. It continuously identifies omissions in the reporting chain, deviations in isolation arrangements, and abnormal work actions during actual operation. Through a multimodal work offset recognition model, it comprehensively analyzes images, sounds, warning zones, and personnel disturbances, and identifies abrupt jumps in local safety parameters based on changes in work rhythm. This allows for the detection of protective failure scenarios before traditional manual inspections, enabling earlier and more granular risk capture. By mapping protective failure scenarios to specific work units and spatially visually annotating them, it automatically integrates responsible persons, approval trajectories, and on-site evidence to form a traceability ledger that can be used for file location and reverse tracing of responsibility chains, making precise risk and responsibility positioning possible. Based on the dynamic updates of the traceability ledger, it automatically refreshes the risk level, prompting methods, and inspection priorities on the monitoring interface, generating directly executable management instructions to the site, thereby improving the efficiency of inspection resource allocation and achieving refined, traceable, and intervention-friendly management of hot work operations. Compared to traditional methods that rely on human experience, this significantly enhances the transparency, timeliness, and responsiveness of hot work operations, helping to reduce the risks of negligent violations and delays, and improving the overall safety level on site. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0016] 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.

[0017] Example 1: Please refer to Figure 1 As shown in the figure, a hot work safety management method based on remote monitoring in an embodiment of the present invention includes the following steps: S1: Construct a basic scenario for the work location, fire source intervention method, and isolation measures, and perform conditional partial order reordering of the reporting link to generate a work access framework that can be recognized by the scheduling engine.

[0018] The process of generating a job admission framework that can be recognized by the scheduling engine in S1 is as follows: The location of the work site is analyzed using on-site geographic information, construction layout maps, and safety isolation points to identify combustible material accumulation areas, confined spaces, and ventilation paths. Existing GIS data, 2D plan layout maps, and 3D modeling data from the construction site are integrated. The work coordinates are accurately located based on positioning tags, ranging points, or on-site base station signals. Spatial overlay analysis technology is used to compare the work point with material storage areas, equipment layout areas, and oil storage areas to automatically identify the location and boundaries of combustible material accumulation areas. By analyzing the enclosed space markings, ventilation shaft layouts, and fan directions in the construction drawings, the confined space range and potential ventilation paths around the work point are identified and marked. Corresponding risk labels are generated based on the identified risk areas. The operation chain of the fire source intervention method is broken down, and a fire source feature group is formed based on the intervention point, direction of action, and heat transfer characteristics. The declared hot work operation type is analyzed, and the corresponding standard operation sequence is retrieved from the operation process library. The operation chain is broken down step by step according to three elements: intervention point, intervention method, and heat transfer path. For example, for electric welding operation, the relative position of the welding machine access point, the welding torch action surface and the welded part is automatically extracted. Combined with the metal thermal conductivity, heat source intensity and spark spatter range, quantifiable heat transfer characteristics are generated. All the decomposed elements are combined into a fire source feature group according to the intervention point position code, the direction of action vector and the estimated heat diffusion model. The consistency of isolation measures is verified, including the continuity of isolation zones, the integrity of fire extinguishing facilities, and the status of locking and tagging. Based on the location and length information of the isolation zones, it is determined whether there are any gaps or insufficient coverage areas. A consistency test is performed by comparing the isolation zone with the standard coverage model. By reading the labels of fire extinguishers, pressure sensors, or image recognition results, it is verified whether the fire extinguishing facilities are complete, within the validity period, and in a usable state. By reading the status of electronic locks, tag identification photos, or RFID lock point tags, it is determined whether locking and tagging have been performed as required. Based on the fire source feature group, operational condition nodes are constructed. A partial order reordering is performed on the approval roles, time sequence, and mutually exclusive conditions in the reporting link. The reordered condition nodes are then used to generate an operational access framework in a structured manner. Core constraints (such as maximum acceptable heat load, combustible material distance limits, and ventilation direction consistency requirements) are extracted from the fire source feature group and mapped to attribute fields of the condition nodes. The processing order, authority boundaries, and mutually exclusive conditions of each approval role in the reporting link are obtained (e.g., some approvals require prior on-site confirmation, and some nodes cannot be handled by the same person). These are then uniformly incorporated into the condition node graph. A partial order relationship reordering algorithm is used to calculate the sequential dependencies, mutually exclusive relationships, and parallel relationships between condition nodes, generating a dynamic approval link that satisfies both regulatory procedures and on-site risk characteristics. The reordered nodes are output in a structured manner, including node ID, triggering conditions, executor roles, a list of prerequisite conditions, and a list of mutually exclusive conditions. This forms an operational access framework that can be directly parsed and scheduled by the scheduling engine, enabling automatic process orchestration and condition-driven triggering.

[0019] S2: Triggers a progressive tracking mechanism for the operation access framework, which reconstructs the operational process changes of hot work activities by calling up the event remnants of historical renovation areas and on-site inspection segments.

[0020] The process of triggering the progressive state tracking mechanism for the job admission framework in S2 is as follows: Based on the trigger conditions set in the work access framework, a status monitoring unit is built for each condition node. After the work access framework is generated, the trigger event type corresponding to each condition node in the framework is automatically read and bound to an independently run monitoring function module, which includes event collection interface, status judgment rules and evidence verification logic. It receives action information uploaded by the reporting system, approval system, on-site inspection terminal and personnel attendance terminal respectively. When the node requires confirmation of on-site isolation arrangement, the monitoring unit will automatically receive photos, videos or inspection records, and check the image integrity, recording time and recording source according to preset requirements. When the node involves approval operation, the monitoring unit will extract the submission action, approval opinion and approval completion time from the approval flow in real time, and check whether it is consistent with the expected role. The monitoring unit can run as a service module, mobile client plugin or lightweight terminal deployed on site to ensure that the data collected from any channel can be identified in real time and converted into node status update trigger input. The monitoring unit continuously records reported actions, approval operations, on-site layout changes, and the on-duty status of workers. Upon receiving any information related to an action, the monitoring unit immediately conducts a preliminary verification of the source, time, and authenticity of the information, and stores the verified information in the event log set in chronological order. For reported actions, it saves the applicant, content summary, and initiation time; for approval operations, it records the approver, operation type, approval opinion, and attached supporting materials; for on-site layout changes, it records image data near the work site, the deployment status of isolation facilities, and confirmation information of inspection personnel; and for on-duty status, it records the sign-in time, sign-in method, and on-site identification results. All records are stored in an unmodifiable manner and archived by the backend in chronological order, while generating a queryable work record chain to ensure that the scheduling engine can call these records at any time to judge the actual completion status of each node. Using a progressive tracking queue, the trigger time, responsible party, and operation content are uniformly transcribed into state evolution fragments. These fragments are then sequentially checked, eliminating duplicate actions and filling in missing nodes. Upon receiving various records, the time, executor, execution content, and supporting evidence are integrated into structured state fragments and added to the tracking queue in the natural order of the workflow. The tracking queue is responsible for verifying the sequential relationship of all fragments, such as whether approval was completed after reporting or whether site setup was completed before hot work. If a logical sequence discrepancy is found, the fragment is marked as abnormal and enters the manual review process. For duplicate reports due to human error, duplicate fragments are automatically filtered out by comparing content, time, and responsible party, retaining only the earliest and most complete record. For steps that should appear in the process but do not, such as not receiving a certain pre-confirmation action, a supplementary reporting reminder is pushed to the corresponding responsible party, and the fragment is automatically filled in upon receipt of the supplementary report. If there is still no response after multiple reminders, the item is marked as a missing node, and the process is prohibited from proceeding to the next stage.

[0021] The process of changing the operational procedures for reorganizing hot work activities in S2 is as follows: The system retrieves event remnants matching the current work area from the historical renovation area database, including equipment renovation records, violation fragments, and accident warning items. Based on the spatial coordinates of the current work area, construction unit identifiers, and work area boundary information, a range query is performed in the historical renovation area database to retrieve all historical records that overlap with or are adjacent to the work area. These historical records include equipment renovation work orders, equipment disassembly and assembly records, violation fragments recorded in previous inspection reports, and previous accident warnings and hazard registration items. A multi-dimensional indexing strategy is used during the retrieval, with time intervals, equipment types, risk levels, and responsible units as search criteria to improve positioning accuracy. The search results are initially sorted by time sequence and risk relevance, and a unified structured remnant record is generated for each record, including the time of occurrence, event summary, associated equipment or structure, type and location of evidence found, responsible person or responsible unit, and historical handling records. Align the event remnant set with the on-site inspection footage according to time tags, and perform image quality assessment to remove blurry segments; extract image segments and records that overlap or are close in time with the historical remnants retrieved from the on-site inspection storage medium, align the two according to the time tags of the images and records to form a preliminary event-image pairing set, and perform quality assessment on each image segment, including image sharpness, exposure, motion blur degree and key content identifiability, and determine whether the image meets the analysis requirements by edge gradient statistics, contrast distribution and inter-frame differences. Image segments that are deemed unqualified are marked and removed, and re-acquisition or on-site reshoot is triggered if necessary. Qualified images are bound to the corresponding historical remnant records, and the image source, shooting equipment, shooting angle and time range are noted in the pairing records; Feature extraction is performed on protective arrangements, operational actions, and surrounding interference factors appearing in the patrol segments to form a sequence of feature nodes. For the quality-assessed images and patrol records, target identification and semantic segmentation are performed to identify isolation zones, warning signs, mobile machinery, welding or cutting activities, personnel postures, and surrounding piles. Descriptive features are extracted from the identified elements, including the type and location of protective facilities, the integrity score of the arrangement, the type and duration of operational actions, the estimated distance of the operational location relative to combustibles, and the presence and intensity of surrounding interference factors. For operational actions, time axis segmentation is also performed to capture the sequence and rhythm changes of actions. For personnel behavior, key indicators such as whether protective equipment is worn and whether the operation is carried out in accordance with regulations are extracted. Each image or record segment is thus converted into a series of feature nodes. The node record includes node type, element description, evidence citation, time interval, and confidence level evaluation. These feature nodes are organized into a sequence according to their appearance order on the time axis to form a fine-grained characterization of on-site operational activities. By combining typical violation patterns from event remnants, similarity clustering is performed on feature node sequences to identify key change points in operational activities and output time-ordered clues of operational process changes. Typical violation patterns annotated by experts from historical remnants are used as a reference sample library. Each pattern in the sample library includes a semantic description of the violation type, a typical feature node sequence, and corresponding handling suggestions and hazard consequences. For the feature node sequences generated during the current inspection, they are compared with patterns in the sample library using similarity metrics. Sequences within a cluster are divided into several clusters based on their similarity. Sequences within a cluster exhibit similar protective defects or abnormal behavior. Through time-series analysis of the clustering results, node segments most closely resembling historical violation patterns during operations are identified. These nodes are labeled as key change points, and for each key point, change clues are generated including the time of occurrence, location, triggering factors, similar historical patterns, and confidence recommendations.

[0022] S3: By using the operation offset recognition model, the system comprehensively analyzes the images, sounds, warning zone markings, and personnel disturbances transmitted from the site. It uses the changes in the operation rhythm depicted by the clues of the operation process as a reference to identify the sudden jumps in local safety parameters and automatically infer the protection failure scenarios that occur.

[0023] The process of comprehensively analyzing the images, sounds, warning zone markings, and personnel disturbances transmitted from the site using the operation offset recognition model in S3 is as follows: The images transmitted from the site are preprocessed according to viewing angle, lighting, and local occlusion. Key fire source areas and work boundary contours are extracted. Noise spectrum decomposition is performed on the sound signals to identify specific frequency bands associated with abnormal impacts and unstable combustion. The brightness, contrast, and shadow areas in the images are dynamically adjusted to make the details around the fire source easier to identify. Perspective correction is performed on the images according to the camera deployment angle to unify the tilted viewing angle to a standard view that is easy to analyze the boundary. For areas with local occlusion caused by personnel, equipment, or smoke, effective images for identification are generated through continuous frame inference, background restoration, and edge completion. Based on this, areas that may generate fire sources are located, and their outer contours, bright spot morphology, and thermal light characteristics are extracted. At the same time, the safety boundary of the work area is identified. For the sound signals, noise spectrum decomposition is performed according to the noise characteristics under different working conditions to remove wind noise, motor noise, and metal friction from the signals to be analyzed. In the purified sound spectrum, frequency bands with abnormal impacts, thermally unstable combustion, and metal bursting are identified and marked as acoustic feature points that may have behavioral deviations or safety risks. The system utilizes a structured recognition model to detect the integrity and continuity of warning zone markings, generating area boundary offsets. It then performs trajectory fitting on personnel position disturbances, analyzing walking paths, dwell times, and potential boundary crossing behaviors. From the processed images, it automatically identifies warning tape, bollards, fences, and ground warning lines. Based on the connectivity, color consistency, and preset positions of these markings in the site layout diagram, it determines whether they are detached, obstructed, moved, or have breaks. It calculates the distance difference, morphological offset, and missing length between the identified areas and the original warning zone to obtain the area boundary offset. For personnel position disturbances, it obtains the relative coordinates of personnel inside and outside the warning zone through human detection and keypoint estimation methods in consecutive frames, and generates walking path curves through trajectory fitting. Combining path turning points, dwell times, and changes in distance to the warning boundary, it determines whether personnel are loitering, approaching high-risk points, suddenly rushing, or crossing the warning zone. Images, sounds, warning zone markers, and personnel disturbances are fused using timestamps and spatial coordinates to form a multimodal input set for offset detection. Based on the acquisition time of each data source, image frames, sound clips, warning zone marker status, and personnel trajectory points are time-aligned, mapping data from different acquisition frequencies onto a synchronized timeline in seconds or milliseconds. Simultaneously, camera locations, microphone placement points, warning zone coordinates, and personnel positioning records are combined to perform coordinate transformations within a unified spatial reference framework, ensuring location-based correspondences between modalities. During fusion, image features, acoustic features, warning zone offsets, and personnel behavior features within each time slice are correlated and analyzed, encapsulated into multimodal input units. Each unit contains source information, spatial location, trigger time, feature summary, and confidence level. All input units are arranged chronologically to form the multimodal input set required for offset detection. The operation offset recognition model further performs event determination, risk identification, and offset quantification, thereby achieving continuous offset detection and dynamic risk identification during on-site operations.

[0024] The process of automatically inferring protection failure scenarios in S3 is as follows: Align the temporal features of the multimodal input set with the clues of changes in the operation process, and calculate the operation rhythm offset at different stages; map the feature events from images, sounds, alert zone status and personnel trajectories onto the same time axis in chronological order, divide them into several stages according to the time window of the event occurrence, so that each stage corresponds to a coherent action segment in the operation process; for each stage, count the key event interval and event density within the stage, and form a baseline rhythm reference by combining the operation rhythm features of the same period in history; compare the rhythm features of the current stage with the baseline reference to measure the rhythm offset, including the acceleration or deceleration of the event occurrence frequency, the advancement or delay of the time of important actions, and the change of action sequence arrangement; Dynamic thresholds are established for the approach speed of heat sources, the range of personnel movement, and the degree of deviation of warning zones. A sudden jump detection method is used to identify discontinuities and rapid spikes in parameter changes. Two reference windows are maintained for each key parameter: a short-term reference window to capture real-time changes on site and a long-term reference window to reflect normal behavior and historical fluctuation range. Based on the comparison results between the short-term and long-term reference windows, the current alarm threshold is automatically adjusted so that the threshold can take into account both instantaneous fluctuations and historical stability on site. For the identification of abnormal jumps, the time-series changes of parameters are smoothed to eliminate random noise. Then, discontinuities and rapid rises are detected on the smoothed curve as jump candidates. When further verifying the candidate jumps, image and sound evidence are retrieved simultaneously to verify the simultaneous occurrence of sudden changes in heat source brightness, equipment vibration or abnormal sounds, and whether personnel trajectories show rapid approach or multiple boundary crossings in a short period of time. The system categorizes leap nodes by risk type and automatically infers protective failure scenarios based on their spatial distribution and change chains. Each confirmed leap node is classified into several risk categories based on its main triggering elements, such as abnormal heat source, personnel approach, breach of isolation measures, and multiple-factor superposition. Each category corresponds to a predefined scenario template, including typical characteristics, possible consequences, and recommended actions. Multiple leap nodes in the same time period or near-spatial locations are linked chronologically to form change chains. The triggering order and causal relationship of events in the chain are analyzed to identify whether a chain reaction caused by a single failure or a complex scenario resulting from the convergence of multiple independent events. Based on chain analysis and comparison with historical typical cases, a description of the protective failure scenario is automatically generated, including the location, triggering points, scope of impact, similar historical cases, and confidence level. Preliminary handling suggestions and priority measures are automatically proposed based on the severity of the scenario. All scenario inference results are output in structured report form, simultaneously linked to video and audio evidence, personnel trajectories, and warning tape offset records, allowing on-site commanders to make immediate judgments and take measures. This information can also serve as a basis for post-event responsibility determination and improvement measure formulation.

[0025] S4: Digitally label the work units and impact range corresponding to the protection failure scenario in the background monitoring interface, integrate the responsible persons, approval trajectories and execution evidence within the scope, and form a hot work traceability ledger for file location and reverse tracing of responsibility links.

[0026] In S4, the process of digitally labeling the work unit and impact range corresponding to the protection failure scenario in the background monitoring interface is as follows: The system maps the scenario types in the protection failure scenarios to the corresponding work units, including fire source units, isolation units, and patrol units. A set of work unit classification rule bases is maintained in the background, including fire source units, isolation units, and patrol units. Each category corresponds to the typical scope of work activities, on-site participating roles, and risk points that should be monitored. When a protection failure scenario is generated, the triggering elements, event links, and objects affected by the scenario are extracted from the scenario description and compared item by item according to the rule base. For example, when the scenario is characterized by a sudden approach of a heat source or a spark deviation, the scenario is classified as a fire source unit. When there is a loose warning tape, a deflected barrier, or insufficient isolation, it is classified as an isolation unit. When personnel frequently stay in the restricted area or enter the dangerous area against the flow of traffic, it is mapped to the patrol unit. The risk range involved in the scenario is expanded by spatial grid to form a visualized impact area; the spatial location of the scenario is obtained from on-site video, positioning devices or control points, and a basic risk area is constructed with the spatial location as the center. The preset spatial grid division scheme is called to divide the on-site area into several grid units of fixed size. Each grid unit represents a calculable spatial segment. According to the scenario type and intensity, the central grid is expanded outward according to certain expansion rules. For example, for fire source scenarios, the grid is expanded first in the direction of flammable material accumulation. For personnel crossing the boundary scenario, the grid is expanded according to the direction of possible movement and range of stay of personnel. The expanded grid set constitutes the visualized impact area of ​​the scenario. The regulatory interface digitally labels work units and affected areas, including risk level, trigger time, and relevant segment indexes. The interface retrieves the mapped work unit information and the spatially expanded affected area, presenting them as layers, markers, or area boxes. For each work unit, the risk level, trigger time, and index number of relevant image or sound segments are labeled near its location, along with a clickable detailed information window. Affected areas are distinguished by different color depths or border styles based on risk levels, ensuring regulators can intuitively see the risk coverage and trends. All labeled content is stored in a backend record table, including label generation time, operation source, and associated scenario number, for retrospective analysis and comparison of historical risk events. The labeling interface supports dynamic updates; when a new scenario occurs or an existing scenario is resolved, the labels automatically refresh, maintaining the real-time accuracy of regulatory information.

[0027] The process of creating a hot work traceability log in S4 for archival location and reverse tracing of responsibility is as follows: The responsible entity information in the digital annotation is matched with the roles and process nodes in the approval system. The responsible entities corresponding to the scenarios are extracted from the digital annotation records, including on-site operators, team leaders, safety administrators, supervisors, and approvers. At the same time, relevant trigger nodes are read, such as the time, location, and corresponding work unit of the risk scenario. The data interface of the back-end approval system is called to obtain the reporting number, approval process structure, and role assignment of each link for this hot work operation, including applicant, reviewer, examiner, and approver. The personnel name, position, certificate number, or identity identifier is used for item-by-item matching. If there are personnel with the same name or who perform duties in different positions, a second comparison is made based on the work unit type and trigger node to ensure that the responsible entity corresponds to a unique role in the approval process. After the matching is completed, each responsible entity is mapped to the corresponding link in the process node. The system structurally integrates the action records, approval timestamps, and execution evidence of the responsible parties to form responsibility chain segments. All responsibility chain segments are then sequentially linked to fill in any missing approval or inspection nodes. Records from various sources are uniformly analyzed, including action segments triggered by on-site cameras, communication content reflected in voice recordings, trajectory data of workers entering and exiting restricted areas, time nodes automatically generated by the approval system, and inspection content in patrol records. Using a timeline alignment method, each record is sorted by occurrence time, and the action subject, operation purpose, execution result, and corresponding evidence documents, such as video segment numbers, image snapshots, or text records, are extracted. Each operation is encapsulated as a responsibility chain segment, containing the responsible party's identity, action type, start time, end time, and verifiable evidence path. All responsibility chain segments are linked to form a continuous operational responsibility chain. For missing links in the chain, such as missing review or inspection nodes, the missing role is inferred from the process template and on-site records, and a blank segment is inserted, simultaneously marked as pending confirmation, ensuring the responsibility chain structure is complete and traceable. A reverse backtracking rule is introduced to trace the responsibility chain backward from the latest situation, locate the responsibility attribution link, and form a hot work operation traceability ledger for file location and reverse backtracking of the responsibility chain. The latest protection failure situation is used as the backtracking starting point. Starting from the responsibility chain segment corresponding to the protection failure situation, the backtracking is carried out in reverse chronological order. According to the preset responsibility classification rules, the key behaviors that caused the situation are retrieved. For example, when the situation is characterized by insufficient isolation, the focus is on querying the inspection records, deployment execution records, and approval confirmation records related to isolation deployment. When the situation is related to personnel crossing the boundary, the on-duty status of the supervisor and the maintenance records of the warning area are checked first. During the backtracking process, the responsibility attribution node is determined according to the risk type of the segment, the effectiveness of the action, and the completeness of the evidence. The key responsible parties that may have caused the situation are marked. All backtracked responsibility nodes, corresponding operations, occurrence time, evidence path, and responsibility explanation are integrated into the hot work operation traceability ledger.

[0028] S5: Based on the dynamic updates of the hot work traceability log, adjust the identification color, prompting method and inspection priority in real time, and generate management instructions for the construction site.

[0029] The process of generating management instructions for construction sites in S5 is as follows: Based on the risk evolution status in the traceability ledger, the risk level of each marked element in the current regulatory interface is dynamically refreshed. The evolution trend information of each risk scenario is read from the traceability ledger, including the start time of risk triggering, the direction of risk spread, changes in scenario type, and the execution status of the corresponding responsibility link. The existing marked elements in the regulatory interface, such as the digital marked positions and current risk levels of fire source units, isolation units, and patrol units, are obtained. According to the preset risk level judgment rules, the risk evolution results in the traceability ledger are compared with the marked elements one by one. For example, when the responsibility link of a certain work unit shows a delay in key operations or a lack of patrol, the risk level of the unit is automatically judged to be upgraded. After the risk level is judged, the level of the marked elements in the regulatory interface is refreshed. The refresh action includes updating the risk color, redrawing the boundary layer, or adjusting the transparency of the layer to ensure that the risk status displayed on the interface is consistent with the actual evolution trend. Different color and prompting methods are applied to labeled elements with different risk levels, including flashing, voice prompts, or vibration prompts. According to the updated risk level, each labeled element is divided into four categories: normal, attention, warning, and severe, and each is configured with different visualization methods. Normal level is displayed in the normal color, attention level is added with a light yellow outline on the edge of the icon, when a labeled element is rated as warning level, its labeled area turns orange and a text prompt box pops up on the right side of the interface, if the risk reaches the severe level, it is displayed in red flashing, and a warning signal is pushed to the on-site management personnel in conjunction with voice broadcast or vibration reminder device. For patrol personnel with portable terminals installed, prompt information is automatically sent to the terminal, so that they can perceive the change of risk level in time during the patrol. Multimodal prompting methods ensure that different risk levels can be quickly identified by on-site personnel and prompt the relevant responsible parties to take immediate countermeasures. The patrol routes are reordered, with high-risk areas highlighted along the patrol path in advance, generating management instructions for the construction site. Based on the updated risk levels, the original link structure of the on-site patrol path is obtained, including fixed patrol points, patrol order, shortest walking distance, and patrol cycle for each point. Elements marked with high risk are prioritized for patrol. By calculating the distance to their destination, the current location of the patrol personnel, and the distribution of obstacles on site, risk points are re-inserted at the beginning of the patrol path to ensure that patrol personnel can reach high-risk areas first. After sorting, the newly generated patrol routes are pushed to the patrol terminals in the form of instructions. The instructions include areas to be inspected first, the arrival path, recommended passages, and corresponding risk warnings. For the on-site broadcast system, a voice version of the management instructions is output simultaneously to guide on-site management personnel to promptly go to key areas to carry out the response. The generated management instructions are used to guide the on-site rapid response to changes in risk and improve the safety management efficiency of hot work operations.

[0030] Example 2: Figure 2 As shown, a hot work safety management system based on remote monitoring includes: Work access module: The location elements, ignition source methods and isolation and protection arrangements for hot work are structured to generate a work access framework; Status tracking module: After the access framework is triggered, retrieve historical event remnants and inspection records to reconstruct clues of changes in the operation process; Risk identification module: Integrates images, sounds, warning zone markings, and personnel disturbances to infer protection failure scenarios based on clues of changes in the work process; Traceability Ledger Module: Digitally marks protection failure scenarios, integrates responsible persons, approval trajectories and execution evidence to form a hot work operation traceability ledger; Command scheduling module: dynamically adjusts and generates management commands based on the updated results of the hot work traceability log.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for safe management of hot work operations based on remote monitoring, characterized in that, Includes the following steps: A preliminary scenario is constructed for the work location, fire source intervention method, and isolation measures. The reporting link is rearranged in a conditional order to generate a work access framework that can be recognized by the scheduling engine. The operation access framework triggers a progressive tracking mechanism, which reconstructs the operation process changes of hot work activities by calling the event remnant set of historical renovation areas and on-site inspection fragments. The operation offset recognition model comprehensively analyzes the images, sounds, warning zone markings and personnel disturbances transmitted from the site. Taking the changes in operation rhythm depicted by the clues of operation process changes as a reference, it identifies the sudden nodes of local safety parameters and automatically infers the protection failure scenarios that occur. The work units and impact range corresponding to the protection failure scenarios are digitally marked in the background monitoring interface. The responsible persons, approval trajectories and execution evidence within the scope are integrated to form a hot work traceability ledger for file location and reverse tracing of responsibility links. Based on the dynamic updates of the hot work traceability log, the identification colors, prompts, and inspection priorities are adjusted in real time to generate management instructions for the construction site.

2. The method for hot work safety management based on remote monitoring according to claim 1, characterized in that, The process of generating a job admission framework that can be recognized by the scheduling engine is as follows: By analyzing the location of the work site using on-site geographic information, construction layout map and safety isolation points, we can identify areas where combustibles accumulate, confined spaces and ventilation paths. The operational chain of fire source intervention methods is broken down, and fire source characteristic groups are formed based on the intervention point, direction of action and heat transfer characteristics. The consistency of the isolation measures deployment was verified, including the continuity of the isolation zone, the integrity of the fire extinguishing facilities, and the status of locks and tags. Based on the fire source characteristic group, the operation condition nodes are constructed. The approval roles, time sequence and mutually exclusive conditions in the reporting link are rearranged in a partial order. The rearranged condition nodes are then used to generate the operation access framework in a structured manner.

3. The method for hot work safety management based on remote monitoring according to claim 2, characterized in that, The process of triggering the progressive state tracking mechanism for the job admission framework is as follows: Based on the triggering conditions set in the job admission framework, a status monitoring unit is built for each condition node; The monitoring unit continuously records reporting actions, approval processes, changes in on-site layout, and the on-site status of workers. By using a progressive tracking queue, the trigger time, responsible party, and operation content are uniformly transcribed into state evolution fragments. The state evolution fragments are then sequentially checked to eliminate duplicate actions and fill in missing nodes.

4. The method for hot work safety management based on remote monitoring according to claim 3, characterized in that, The process of reorganizing the hot work operation process changes is as follows: Retrieve event remnants that match the current work area from the historical modification area database, including equipment modification records, violation fragments, and accident warning items; Align the event remnant set with the on-site inspection footage by time tag, and perform image quality assessment to remove blurry segments; Feature extraction is performed on the protective arrangements, operational actions, and surrounding interference factors that appear in the patrol segments to form a sequence of feature nodes; By combining typical violation patterns in the event residue set, similarity clustering is performed on the feature node sequence to identify key change points in the operation activities and output clues of operation process changes sorted by time.

5. A method for safe management of hot work operations based on remote monitoring according to claim 4, characterized in that, The process of comprehensively analyzing the images, sounds, warning zone markings, and personnel disturbances transmitted from the site using the operation offset recognition model is as follows: The images transmitted from the scene are preprocessed according to the viewpoint, lighting and local occlusion, the key fire source area and the operation boundary contour are extracted, and the sound signal is decomposed into noise spectrum to identify specific frequency bands associated with abnormal knocking and combustion instability. The integrity and continuity of warning zone markings are detected using a structured recognition model, the area boundary offset is generated, the trajectory of personnel position disturbance is fitted, and the walking path, dwell time and potential boundary crossing behavior are analyzed. Images, sounds, warning zone markers, and personnel disturbances are fused together with timestamps and spatial coordinates to form a multimodal input set for offset detection.

6. A method for safe management of hot work operations based on remote monitoring according to claim 5, characterized in that, The process of automatically inferring potential protection failure scenarios is as follows: Align the temporal features in the multimodal input set with the cues of changes in the work process, and calculate the work rhythm offset at different stages; Dynamic thresholds are established for the speed of heat source approach, the range of personnel movement, and the degree of warning zone deviation. A sudden jump detection method is used to identify discontinuities and rapid spikes in parameter changes. The system categorizes leap nodes by risk type and automatically infers potential protection failure scenarios based on the spatial distribution of leap nodes and changing links.

7. A method for safe management of hot work operations based on remote monitoring according to claim 6, characterized in that, The process of digitally labeling the work units and impact range corresponding to protection failure scenarios in the background monitoring interface is as follows: Map the scenario types in the protection failure scenario to the corresponding work units, including fire source unit, isolation unit and patrol unit; The scope of risks involved in the scenario is expanded using a spatial grid to form a visualized area of ​​impact; The work units and affected areas are digitally labeled in the regulatory interface, and the labeling content includes risk level, trigger time and relevant fragment index.

8. A method for hot work safety management based on remote monitoring according to claim 7, characterized in that, The process of creating a hot work operation traceability log for archival location and reverse tracing of responsibility is as follows: The information of the responsible parties in the digital annotations is matched with the roles and process nodes in the approval system; The action records, approval timestamps, and execution evidence of the responsible parties are structured and integrated to form responsibility chain segments. All responsibility chain segments are then strung together in chronological order to fill in any missing approval or inspection nodes. By introducing reverse backtracking rules, the responsibility chain is traced backward from the latest situation to locate the link of responsibility and form a hot work operation traceability ledger for file location and reverse backtracking of the responsibility chain.

9. A method for safe management of hot work operations based on remote monitoring according to claim 8, characterized in that, The process of generating management instructions for the construction site is as follows: Based on the risk evolution status in the traceability ledger, the risk level of each marked element in the current regulatory interface is dynamically refreshed. Different color and prompting methods are applied to the labeled elements of different risk levels, including flashing, voice prompts or vibration prompts; The patrol routes are reordered, and high-risk areas are highlighted in advance along the patrol path, generating management instructions for the construction site.

10. A hot work safety management system based on remote monitoring, applied to the hot work safety management method based on remote monitoring as described in any one of claims 1-9, characterized in that, include: Work access module: The location elements, ignition source methods and isolation and protection arrangements for hot work are structured to generate a work access framework; Status tracking module: After the access framework is triggered, retrieve historical event remnants and inspection records to reconstruct clues of changes in the operation process; Risk identification module: Integrates images, sounds, warning zone markings, and personnel disturbances to infer protection failure scenarios based on clues of changes in the work process; Traceability Ledger Module: Digitally marks protection failure scenarios, integrates responsible persons, approval trajectories and execution evidence to form a hot work operation traceability ledger; Command scheduling module: dynamically adjusts and generates management commands based on the updated results of the hot work traceability log.

Images