A management system and method for identifying violations and safety hazards at construction sites.

Abstract

This invention, entitled "A Management System and Method for Irregularities and Safety Hazards at Construction Sites," belongs to the field of construction safety management technology. The technical problems it addresses include the high safety risks, low efficiency and high rate of missed detection associated with traditional manual inspections, lack of closed-loop hazard handling, and difficulties in data traceability. It also addresses the limitations of existing inspection robots in adapting to unstructured terrain and the lack of customized AI recognition algorithms. The key technical solution involves using a quadruped robot dog inspection terminal, equipped with an edge computing module featuring a built-in customized AI recognition model for specific engineering scenarios. This is combined with a cloud management platform, a mobile app, and a data storage module. Data interaction is achieved through 4G / 5G wireless communication, constructing a closed-loop management system that integrates with the engineering BIM model to achieve visualized hazard management and full lifecycle data traceability.

Description

Technical Field

[0001] This invention belongs to the field of engineering construction safety management technology, specifically relating to a management system and method for identifying violations and safety hazards at engineering construction sites. Background Technology

[0002] Pumped storage project construction sites are complex environments, encompassing various special scenarios such as underground cavern complexes, steep slopes, blasting areas, and electrical work areas. Workers are dispersed and there are numerous high-risk operations involving heights, live electrical work, and hazardous materials. Traditional manual inspection methods are no longer sufficient to meet the actual needs of safety management and have many limitations. (1) High safety risks of manual inspection: There are often safety hazards such as toxic gas accumulation, falling rocks, and collapse in the construction area. When inspection personnel enter such areas, they face extremely high personal safety risks. Working in a high-humidity, low-oxygen, and dusty environment for a long time can easily cause occupational diseases. At the same time, with the aging of the population, there is an increasing shortage of young laborers willing to engage in arduous underground work.

[0003] (2) Low efficiency in identifying violations and high rate of missed detection: Manual inspections rely on the subjective experience and physical condition of the inspectors, which makes it difficult to identify unsafe behaviors such as workers not wearing safety helmets or smoking in violation of regulations, as well as safety hazards that are more concealed; it is difficult to achieve full coverage of the construction site around the clock, and there are a lot of blind spots in supervision; especially in special working conditions such as dark light, smoke, and blasting warning zones, the difficulty of identification increases significantly.

[0004] (3) Lack of closed-loop management of hidden dangers: The existing inspection mode can only realize the one-way process of "discovering hidden dangers - reporting hidden dangers", lacking full-process tracking and control of hidden danger rectification, review and cancellation. It is easy to have the problem of "no one to handle the hidden dangers after they are reported, and no one to review them after they are handled", resulting in the long-term existence of safety hazards.

[0005] (4) Difficulty in data traceability: The data recorded by manual inspection is mostly paper or fragmented electronic records, which are difficult to link with the engineering BIM model and digital twin system. It is impossible to form a complete data chain for hazard identification, handling and verification. When quality and safety problems occur later, it is difficult to define responsibility and there is a lack of traceable evidence chain.

[0006] While some engineering inspection robots have emerged in the existing technology, they suffer from the following shortcomings: most are wheeled or track-based robots, making it difficult to adapt to unstructured terrain such as gravel, mud, and water accumulation at pumped storage engineering construction sites; at the same time, existing robots mostly only have basic video acquisition functions and lack AI intelligent recognition algorithms customized for engineering scenarios, making it impossible to accurately identify violations; and a complete closed-loop management system for potential hazards has not been established, making it difficult to fundamentally solve the pain points of traditional inspection models.

[0007] In view of this, the present invention is hereby proposed. Summary of the Invention

[0008] To address the aforementioned technical problems in the existing technology, this invention provides a management system and method for identifying violations and safety hazards at construction sites. This system solves the problems of high safety risks, low efficiency in identifying violations, lack of closed-loop management of hazards, and difficulty in data traceability associated with traditional manual inspections. It enables real-time and accurate identification of violations at construction sites and closed-loop management of safety hazards throughout the entire process, thereby improving the intelligence and precision of construction safety management and ensuring construction safety.

[0009] To achieve the above objectives, the technical solution of the present invention is as follows: Firstly, a management system for violations and safety hazards at construction sites includes a robot dog inspection terminal, an edge computing module, a cloud management platform, a mobile APP, and a data storage module. The robot dog inspection terminal integrates a multimodal perception module, an audible and visual alarm module, a positioning module, and a communication module. The robot dog inspection terminal collects environmental data and worker behavior data from the construction site, receives control commands from the cloud management platform, and performs inspection, alarm, and verification tasks. The edge computing module is mounted on the robot dog inspection terminal. The edge computing module has a built-in customized AI recognition model for engineering scenarios. The edge computing module processes the multi-source data collected by the robot dog inspection terminal in real time, completes the identification of violations and the investigation of safety hazards, generates identification results and preliminary alarm information, and uploads key data to the cloud management platform. The cloud management platform receives the identification results, alarm information and inspection data uploaded by the edge computing module. The cloud management platform classifies and manages violations and safety hazards, issues alarms at different levels, assigns tasks, tracks rectification and conducts data statistical analysis. At the same time, it links with the engineering BIM model and digital twin system for data linkage. The mobile app synchronizes data with the cloud management platform in real time. The mobile app receives alarm information and rectification tasks pushed by the cloud management platform, provides feedback on the rectification status, and initiates a review application. The data storage module stores the raw data collected by the robot dog inspection terminal, the recognition results generated by the edge computing module, the control data of the cloud management platform, and the closed-loop processing records of the entire process.

[0010] Furthermore, the robot dog inspection terminal has adaptive gait adjustment capability, and its protection level is not lower than IP67 and it has explosion-proof capability; the positioning module adopts SLAM and BIM fusion navigation technology, combined with IMU inertial measurement unit, and the positioning error of the positioning module is not greater than 10cm.

[0011] Furthermore, the multimodal sensing module includes a 4K low-light industrial camera, an infrared thermal imaging probe, a millimeter-wave sensor, and an environmental sensor; The 4K low-light industrial camera captures video streams of worker behavior and images of equipment status at the construction site; The infrared thermal imaging probe captures hidden heat sources at the construction site. The millimeter-wave sensor detects the distance between the workers and the hazardous area; The environmental sensors collect data on temperature, humidity, dust concentration, and toxic gas concentration at the construction site.

[0012] Furthermore, the edge computing module is equipped with an AI inference chip; The edge computing module preprocesses the data collected by the multimodal perception module, including adaptive histogram equalization and median filtering; the edge computing module uses online distillation technology to transfer cloud model knowledge to a lightweight model at the edge.

[0013] Furthermore, the customized AI recognition model for engineering scenarios is trained based on deep learning algorithms and combined with labeled data from the engineering construction site. The customized AI recognition model for engineering scenarios can identify behaviors such as workers not wearing safety helmets, smoking in violation of regulations, entering high-risk work areas in violation of regulations, not wearing safety ropes when working at heights, and operating electrical equipment in violation of regulations, as well as the conditions of electrical equipment overheating, excessive toxic gases, and loose rocks.

[0014] Furthermore, the cloud management platform includes a closed-loop hazard management module; The hidden danger closed-loop management module performs secondary verification on the received preliminary alarm information, automatically dispatches rectification tasks to the mobile APP of the corresponding management personnel according to the alarm risk level, controls the robot dog inspection terminal to go to the location of the hidden danger to perform the review task, and automatically cancels the number of hidden dangers that pass the review.

[0015] Furthermore, the cloud management platform also includes an inspection management module; The inspection management module generates inspection paths based on the BIM model of the engineering site and electronic fences, supports adjusting the inspection frequency of different areas, remotely controls the inspection status of the robot dog inspection terminal, and obtains inspection progress, location information and real-time video stream in real time.

[0016] Furthermore, the data storage module adopts a combination of cloud-based distributed storage and local caching; The data storage module encrypts the stored data, and the storage period of the data storage module is no less than the construction period plus 5 years; the data storage module has a data backup function.

[0017] Furthermore, the communication module adopts 4G and 5G dual-mode communication, and the transmission delay of the communication module is no more than 50ms; The communication module uploads the identification results, alarm information and inspection data to the cloud management platform, and at the same time receives inspection path adjustment and task assignment control instructions issued by the cloud management platform.

[0018] Secondly, a method for managing violations and safety hazards at construction sites, applied to the construction site violation and safety hazard management system described in any of the preceding paragraphs, includes: S1. Start the robot dog inspection terminal through the cloud management platform or mobile APP, select the preset or custom inspection path, and set the inspection frequency. S2. The robot dog inspection terminal moves autonomously along the inspection path. The positioning module obtains location information in real time and associates it with the BIM model. The multimodal perception module synchronously collects images, videos and environmental parameter data of the construction site and transmits them to the edge computing module. S3, the edge computing module preprocesses the collected data, uses a customized AI recognition model for engineering scenarios to identify violations and investigate safety hazards, and generates recognition results and preliminary alarm information; S4. The edge computing module uploads the recognition results, raw data and preliminary alarm information to the cloud management platform, and at the same time triggers the sound and light alarm of the robot dog inspection terminal. S5, the cloud management platform performs secondary verification of alarm information and assigns rectification tasks to the corresponding managers' mobile apps according to the alarm risk level; S6. Managers receive rectification tasks via mobile app, go to the site to handle them and report the rectification progress, and upload rectification materials before initiating a review application. S7, the cloud management platform controls the robot dog inspection terminal to go to the location of the hidden danger for secondary inspection, and verifies the rectification effect through the customized AI recognition model of the engineering scenario and generates a review report; S8 and the cloud management platform review and verify reports, automatically canceling the numbers of hazard reports that pass the verification, and the data storage module stores the entire process of handling records and associates them with the BIM model.

[0019] The beneficial effects of this invention are as follows: (1) A quadruped robot dog is used as the inspection terminal. It has adaptive gait adjustment capability, high protection level and good explosion-proof performance. It can be adapted to unstructured extreme scenarios such as underground caverns, piles of gravel and water accumulation pits in pumped storage projects. It breaks through the scene adaptation limitations of traditional wheeled and track-type robots, freeing operators from heavy and dangerous inspection environments and effectively avoiding personal safety risks faced by inspection personnel.

[0020] (2) Integrate multimodal perception modules and combine data preprocessing technology to solve the problems of low recognition accuracy and high false alarm rate in extreme environments; construct customized AI recognition models for engineering scenarios, combine the characteristics of violations and safety hazards in pumped storage projects, and continuously optimize the models through technologies such as online distillation to achieve real-time and accurate recognition of violations and safety hazards. The recognition efficiency is significantly improved compared with manual inspection, the missed detection rate is greatly reduced, and the inspection of the entire construction site is achieved in all areas and all weather conditions, eliminating blind spots in supervision.

[0021] (3) Construct a closed-loop management system for the entire process of “perception-identification-alarm-dispatch-rectification-review-cancellation”, combining robot dog inspection, AI identification, cloud control and mobile terminal linkage to achieve full-process tracking and control of violations and safety hazards, solve the pain point of traditional inspection “only discovers, does not deal with”, ensure that hidden dangers are eliminated in time, and effectively reduce the incidence of safety accidents.

[0022] (4) Real-time linkage with the engineering BIM model and digital twin system is realized, and the location of hidden dangers, identification results and closed-loop handling records are accurately superimposed on the BIM model to form an unalterable safety management data archive, realize the visual management of hidden dangers and the full life cycle data traceability, and provide reliable support for responsibility definition and safety decision-making.

[0023] (5) This system can be adapted to various complex engineering construction sites such as pumped storage, water conservancy, transportation, and mining. It can be deployed and used without large-scale modification and has good versatility and promotion. Attached Figure Description

[0024] Figure 1 An architecture diagram of the management system for violations and safety hazards at engineering construction sites provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the robot dog inspection terminal provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the workflow of an edge computing module provided in an embodiment of the present invention; Figure 4 A schematic diagram of the closed-loop management process for potential hazards provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the deployment at a pumped storage project construction site, provided by an embodiment of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0026] It should be noted that, unless otherwise specifically stated, the relative arrangement and numerical expressions of the components and steps described in these embodiments should not be construed as limiting the scope of the invention.

[0027] The following description of exemplary embodiments is merely illustrative and is not intended to limit the invention or its application or use in any way. Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but where applicable, such techniques, methods, and apparatus should be considered part of this specification.

[0028] Example 1 See Figure 1 , Figure 1 This is an architecture diagram of a management system for violations and safety hazards at construction sites proposed in this invention. It includes a robot dog inspection terminal, an edge computing module, a cloud management platform, a mobile app, and a data storage module. The modules interact with each other via 4G / 5G wireless communication. Specifically, it includes: M1, Robot Dog Inspection Terminal like Figure 2 As shown, the robot dog inspection terminal is a quadruped robot with adaptive gait adjustment capabilities, making it adaptable to unstructured terrains such as piles of gravel, puddles, and steel mesh. The robot dog inspection terminal has a protection rating of at least IP67, is explosion-proof, and can withstand extreme environments such as high humidity, dust, low light, and toxic gases in underground caverns of pumped storage projects.

[0029] The robot dog inspection terminal integrates a multimodal perception module, an audible and visual alarm module, a positioning module, and a communication module.

[0030] M11, a multimodal sensing module, includes a 4K low-light industrial camera, an infrared thermal imaging probe, a millimeter-wave sensor, and an environmental sensor, among which: 4K low-light industrial cameras capture video streams of workers' behavior and images of equipment status at construction sites. They are adaptable to complex lighting conditions such as low light and backlight, and can capture details such as workers not wearing safety helmets and smoking in violation of regulations. Infrared thermal imaging probes capture hidden heat sources at construction sites, helping to identify violations and safety hazards such as overheating of electrical equipment in low-light and smoky environments. Millimeter-wave sensors assist in detecting the distance between workers and hazardous areas; Environmental sensors collect environmental parameters such as temperature, humidity, dust concentration, and toxic gas concentration at the construction site, providing environmental references for identifying violations and investigating potential hazards.

[0031] The M12 positioning module adopts SLAM+BIM fusion navigation technology and combines an IMU inertial measurement unit to achieve accurate positioning in the absence of GPS, with a positioning error of ≤10cm. It can obtain the location information of the robot dog inspection terminal in real time and associate it with the BIM model to achieve accurate marking of the location of hidden dangers.

[0032] M13, the sound and light alarm module, immediately triggers an on-site sound and light alarm when a violation or safety hazard is detected, with a sound intensity ≥85dB and flashing warning lights.

[0033] The M14 communication module adopts 4G / 5G dual-mode communication, supports high-speed data transmission, and has a transmission latency of ≤50ms. It uploads the identification results, alarm information and inspection data to the cloud management platform in real time, and at the same time receives control commands such as inspection path adjustment and task assignment issued by the cloud management platform.

[0034] M2, Edge Computing Module like Figure 3 As shown, the edge computing module is installed on the robot dog inspection terminal, equipped with an AI inference chip and a built-in customized AI recognition model for engineering scenarios. This model is based on deep learning algorithms such as YOLOv8 object detection algorithm, CNN convolutional neural network and Transformer visual feature extraction algorithm, combined with the characteristics of violations and safety hazards at the pumped storage project construction site. It was trained and optimized through a large amount of labeled data from the engineering site and has the following functions: Violation Identification: Identifies common violations at construction sites, including workers not wearing safety helmets, smoking in violation of regulations, entering high-risk work areas without permission, working at heights without safety ropes, and operating electrical equipment in violation of regulations, with an accuracy rate of ≥95%. Safety Hazard Investigation: Combining data collected by the multimodal sensing module, the system investigates equipment hazards, environmental hazards, and site hazards at the construction site. Equipment hazards include damaged electrical wiring and equipment overheating; environmental hazards include excessive levels of toxic gases and high dust concentrations; and site hazards include loose rocks and water accumulation blocking passageways. The hazard identification accuracy rate is ≥90%. Data preprocessing and optimization: The images, videos, environmental parameters and other data collected by the multimodal perception module are preprocessed. The interference of dust, water mist and jitter is removed by using adaptive histogram equalization, median filtering and other techniques. At the same time, online distillation technology is used to transfer the knowledge of the complex cloud model to the lightweight edge model. Preliminary alarm judgment: Based on the identification results and combined with the preset violation behavior and safety hazard classification standards, preliminary alarm information is generated, and the alarm type, risk level, location and timestamp are marked. At the same time, the identification results, raw data and preliminary alarm information are uploaded to the cloud management platform. If a major hazard occurs, the audible and visual alarm of the robot dog inspection terminal is directly triggered.

[0035] M3, Cloud Management Platform The cloud management platform receives the identification results, alarm information and inspection data uploaded by the edge computing module, classifies and manages violations and safety hazards, issues alarms at different levels, assigns tasks, tracks rectification and conducts data statistical analysis, and also links data with the engineering BIM model and digital twin system.

[0036] like Figure 4 As shown, the cloud management platform includes a hazard closed-loop management module, an inspection management module, an AI recognition management module, a data statistics and analysis module, an access control module, and a data linkage module.

[0037] M31, the hidden danger closed-loop management module, performs secondary verification on the received initial alarm information, automatically dispatches rectification tasks to the corresponding management personnel's mobile APP according to the alarm risk level, controls the robot dog inspection terminal to go to the location of the hidden danger to perform the verification task, and automatically cancels the number of hidden dangers that pass the verification. The specific process is as follows: Alarm Triggering: Receives preliminary alarm information uploaded by the edge computing module, and performs secondary verification of the alarm information in combination with environmental parameters and on-site scenarios to confirm the validity of the alarm and avoid false alarms; Tiered task assignment: Based on the alarm risk level (general, significant, major) and combined with the actual control requirements of the pumped storage project construction site, clear quantitative grading standards are established. Rectification tasks are automatically assigned to managers in the corresponding areas, and alarm details, hazard locations, and handling suggestions are simultaneously pushed to the managers' mobile app. Specific quantitative standards and task assignment rules are as follows: The quantitative standards for alarm risk levels include: General alarms: The quantitative standard is a single minor hazard / violation that does not directly threaten personnel safety, does not affect the normal operation of equipment, and has no risk of escalation. Specifically, this includes: workers not wearing safety helmets (single person, single incident), minor water accumulation obstructing passage (area < 5㎡, no electrical equipment nearby), minor dust concentration exceeding the standard (exceedance < 20%, duration < 10 minutes), and minor dust accumulation on equipment (not affecting equipment operation). The hazard identification confidence level is 90%-92%.

[0038] Major Alarm: The quantitative standard is a single serious hazard / violation, or multiple minor hazards overlapping, which may indirectly threaten personnel safety, affect the normal operation of equipment, and pose a potential risk of expansion. Specific examples include: unauthorized entry into high-risk work areas (single person, single time, no contact with hazardous equipment), working at height without a safety rope (single person, single time, working height < 5 meters), minor damage to electrical wiring (no exposed copper core, no leakage), slight exceedance of toxic gas standards (exceeding standards by 20%-50%, duration 10-30 minutes), and slightly loose loose rocks (no risk of falling, volume < 0.5 m³). Hazard identification confidence level is 92%-95%.

[0039] Major Alarm: Quantified as a serious hazard / violation, directly threatening personnel safety, potentially causing equipment downtime or safety accidents, and showing an escalating trend. Specific examples include: smoking in violation of regulations (within hot work areas), improper operation of electrical equipment (leading to abnormal equipment operation or leakage), working at height without a safety rope (working height ≥ 5 meters), severely damaged electrical wiring (exposed copper core, with obvious leakage), severely excessive levels of toxic gases (exceeding the standard by ≥ 50%, duration ≥ 30 minutes), severely loose rocks (potential to fall, volume ≥ 0.5 m³), ​​and equipment overheating (temperature exceeding rated value by 30% or more). Hazard identification confidence level ≥ 95%.

[0040] During the task dispatch process, the mobile app synchronizes the task progress in real time (pending receipt, received, under verification, under rectification, completed, reviewed). If the task is not handled within the quantified timeframe, the system automatically triggers a secondary reminder. If the task is not handled within the specified time, it will be escalated to a higher level of management to ensure closed-loop management of potential hazards. Rectification Tracking: After receiving the rectification task, the management personnel can provide feedback on the rectification progress and upload rectification photos / videos through the mobile APP. The cloud management platform tracks the rectification status in real time and reminds and supervises tasks that have not been rectified by the deadline. Verification and Review: After rectification is completed, the management personnel initiate a verification application. The cloud management platform controls the robot dog inspection terminal to go to the location of the hidden danger for a second inspection. The AI ​​recognition model is used to verify the rectification effect and generate a verification report. Hazard cancellation: Once the review is passed, the hazard is automatically cancelled, forming a complete closed-loop handling record; if the review fails, the rectification task is reassigned until the hazard is completely eliminated.

[0041] The M32 inspection management module, based on the engineering site BIM model and electronic fence, automatically generates inspection paths with "priority to key areas + full coverage". It supports manual adjustment of the inspection frequency of key areas such as high-altitude work areas, blasting warning zones, and power distribution rooms. It can remotely control the inspection status of the robot dog inspection terminal, view the inspection progress, location information and real-time video stream in real time, and supports dynamic adjustment of the inspection path.

[0042] The M33 AI recognition and management module establishes a stable data link through the TCP / IP network protocol and the MQTT message push protocol, and uses an encrypted HTTP / HTTPS interface to conduct bidirectional data interaction with the edge computing module. The AI ​​recognition and management module receives structured data such as recognition results, alarm information, location coordinates and timestamps uploaded by the edge computing module in real time, as well as unstructured data such as images and videos. It completes parsing and verification through JSON data format, classifies and archives violations and security risks, and supports multi-condition combination retrieval and statistical analysis by alarm type, risk level, occurrence area, occurrence time and other dimensions.

[0043] Inter-module support for breakpoint resumption and data verification mechanisms ensures data transmission integrity and reliability in weak network environments. Simultaneously, the AI ​​recognition management module manages AI recognition models, performs version updates, incremental updates, and iterative optimizations via an OTA remote upgrade channel. Combining long-tail data from construction sites, and employing a horizontal federated learning and parameter aggregation framework, it uploads only model gradients and updated parameters, without uploading original site data. This achieves continuous improvement in multi-terminal collaborative training and model generalization capabilities without disclosing site data.

[0044] The M34 data statistics and analysis module statistically analyzes data such as the types, frequency, and handling efficiency of violations and safety hazards at the construction site, generating visual reports to provide data support for construction safety management decisions. At the same time, it combines historical data to analyze the patterns of hazard occurrence, provide early warnings of high-risk areas and violations, and achieve proactive prevention and control.

[0045] The M35 access control module assigns different operating permissions based on the administrator's position and responsibilities, ensuring the security and standardization of system operations; it also supports multi-user collaborative operation, enabling a clear division of management responsibilities.

[0046] M36, the data linkage module, is linked in real time with the engineering BIM model and digital twin system, accurately overlaying the location of hazards, identification results, and closed-loop handling records onto the BIM model to achieve visualized management of hazards and full life-cycle data traceability.

[0047] M4, Mobile App The mobile app synchronizes data with the cloud management platform in real time, receives alarm information and rectification tasks pushed by the cloud management platform, provides feedback on rectification status, and initiates review requests. The mobile app also supports real-time viewing of the robot dog inspection terminal's location and video stream, querying historical records and closed-loop handling records, reporting abnormal situations at the construction site, and initiating inspection requests. Specifically, this includes: Alarm notifications: Receive alarm information on violations and security risks pushed by the cloud management platform, including alarm type, risk level, location of occurrence, on-site photos / videos and other details; Task Management: Receive rectification tasks and review tasks, view task details, provide feedback on task progress, and upload photos / videos related to rectification and review; Real-time viewing: View the inspection location and real-time video stream of the robot dog inspection terminal in real time to understand the real-time situation at the construction site; Data Query: Query historical records of violations and safety hazards, as well as closed-loop handling records, supporting searches by time, region, type, and other dimensions; Feedback: Managers can use the app to report any abnormalities at the construction site, initiate inspection requests, and submit rectification applications.

[0048] M5, data storage module The data storage module employs a combination of cloud-based distributed storage and local caching. The local cache stores recently collected data from the robot dog's inspection terminal, automatically synchronizing it to the cloud upon network recovery. Cloud storage encrypts the data, with a storage period of no less than the project construction period plus 5 years, and includes data backup functionality.

[0049] The data storage module stores the raw data collected by the robot dog inspection terminal, the recognition results generated by the edge computing module, the control data of the cloud management platform, and the closed-loop handling records of the entire process.

[0050] like Figure 5 As shown, the deployment steps of this system at the pumped storage project construction site are as follows: Robot dog inspection terminal deployment: Based on the division of the construction site area, deploy several quadruped robot dog inspection terminals, focusing on covering high-risk areas such as underground cavern groups, blasting operation areas, power distribution rooms, and high-altitude operation faces; complete environmental adaptability debugging and multimodal perception module calibration.

[0051] Edge computing module deployment: An edge computing module is installed on each robot dog inspection terminal, a customized AI recognition model for engineering scenarios is installed, and a labeled dataset of violations and safety hazards at the pumped storage engineering construction site is imported to complete on-site debugging and optimization.

[0052] Deployment of cloud management platform: Build a cloud management platform and complete communication debugging with robot dog inspection terminal and mobile APP; import the BIM model and electronic fence data of the project site, set inspection path, frequency and alarm classification standards; configure data storage module to realize encrypted storage and backup of data; assign management personnel permissions and clarify the management responsibilities of each position.

[0053] Mobile App Deployment: Administrators install the mobile app, complete the binding with the cloud management platform, and debug functions such as alarm notifications, task reception, and data query to ensure data synchronization between the mobile app and the cloud.

[0054] Example 2 This embodiment provides a method for managing violations and safety hazards at construction sites, including the following steps: S1. Start the robot dog inspection terminal through the cloud management platform or mobile APP, select the preset or custom inspection path, and set the inspection frequency. S2. The robot dog inspection terminal moves autonomously along the inspection path. The positioning module obtains location information in real time and associates it with the BIM model. The multimodal perception module synchronously collects images, videos and environmental parameter data of the construction site and transmits them to the edge computing module. S3, the edge computing module preprocesses the collected data, uses a customized AI recognition model for engineering scenarios to identify violations and investigate safety hazards, and generates recognition results and preliminary alarm information; S4. The edge computing module uploads the recognition results, raw data and preliminary alarm information to the cloud management platform, and at the same time triggers the sound and light alarm of the robot dog inspection terminal. S5, the cloud management platform performs secondary verification of alarm information and assigns rectification tasks to the corresponding managers' mobile apps according to the alarm risk level; S6. Managers receive rectification tasks via mobile app, go to the site to handle them and report the rectification progress, and upload rectification materials before initiating a review application. S7, the cloud management platform controls the robot dog inspection terminal to go to the location of the hidden danger for secondary inspection, and verifies the rectification effect through the customized AI recognition model of the engineering scenario and generates a review report; S8 and the cloud management platform review and verify reports, automatically canceling the numbers of hazard reports that pass the verification, and the data storage module stores the entire process of handling records and associates them with the BIM model.

[0055] Example 3 This embodiment uses two typical scenarios from pumped storage project construction sites to illustrate the actual operation process and closed-loop management effect of this system; specifically including: B1. Scenario for identifying and handling violations of workers not wearing safety helmets The robotic dog inspection terminal performs an inspection task every hour according to the preset inspection path of the underground cavern. When it moves to area 3 of the underground cavern, the 4K low-light industrial camera captures real-time video streams of the on-site workers and transmits them to the edge computing module. The edge computing module analyzes the video frames through an engineering scenario-customized AI recognition model and identifies one worker who is not wearing a safety helmet, with an accuracy rate of 96%.

[0056] The edge computing module immediately generates preliminary alarm information, indicating the alarm type as "not wearing a safety helmet," the risk level as "moderate," the location as "underground cavern area 3," and the corresponding timestamp. Simultaneously, it triggers the audible and visual alarm on the robot dog inspection terminal, with the sound intensity reaching 85dB and warning lights flashing, reminding on-site personnel to immediately correct the violation. The edge computing module also simultaneously uploads the identification results, original on-site photos, and preliminary alarm information to the cloud management platform.

[0057] After receiving the alarm information, the cloud management platform performs a secondary verification by combining the normal environmental parameters collected by the environmental sensors. If the alarm is confirmed to be valid, the platform automatically assigns the rectification task to the safety management personnel in the underground cavern area and pushes the alarm details, the location of the hazard, and the handling suggestions to their mobile APP.

[0058] After receiving the task via a mobile app, management personnel arrive at the site within 5 minutes, urge workers to wear safety helmets correctly, take photos of the rectified site using the mobile app and upload them, and then initiate a review request. Upon receiving the review request, the cloud management platform automatically dispatches the nearest robot dog inspection terminal to area 3 of the underground cavern for a secondary inspection.

[0059] After the robot dog inspection terminal reaches the designated location, the 4K low-light industrial camera recaptures images of the site. The edge computing module uses an AI recognition model to confirm that all workers are wearing safety helmets correctly, generates a verification report, and uploads it to the cloud management platform. Once approved by the cloud management platform, the hazard is automatically removed from the list. The complete handling process record, on-site photos, and verification report are all encrypted and stored in the data storage module, precisely linked to the corresponding location in the BIM model, supporting subsequent traceability and query.

[0060] B2. Closed-loop management scenario for electrical circuit overheating safety hazards When the robot dog inspection terminal was performing an inspection task in the power distribution room area, the infrared thermal imaging probe detected an abnormally high temperature at a line joint in a low-voltage distribution cabinet, reaching 85°C. At the same time, the environmental sensors collected data showing that the temperature and humidity at the site were normal and the dust concentration met the standards. The edge computing module combined multimodal data for comprehensive analysis and identified it as a "high risk" safety hazard of "electrical equipment overheating".

[0061] The edge computing module immediately generates preliminary alarm information, marking the type of hazard, risk level, specific location and timestamp, triggering the audible and visual alarm of the robot dog inspection terminal, and simultaneously uploading the infrared thermal imaging image, environmental parameters and alarm information to the cloud management platform.

[0062] After verifying the alarm's validity, the cloud management platform automatically assigns the rectification task to electrical management personnel, pushing the suggestion to "immediately disconnect the power, check the tightness of the line joints, and investigate poor contact issues." Upon receiving the task, the electrical management personnel take their tools to the power distribution room. After disconnecting the power, they find that the line joints have become loose due to long-term vibration, leading to increased contact resistance and overheating. They then retighten the joints.

[0063] After rectification was completed, management personnel uploaded videos of the rectification process and photos of the rectified site via a mobile app, initiating a review request. The cloud management platform dispatched a robot dog inspection terminal to the power distribution room for review. The infrared thermal imaging probe re-inspected the line joint, showing that the temperature had returned to the normal range of 32℃. The edge computing module generated a review approval report and uploaded it to the cloud.

[0064] Once approved by the cloud management platform, the hazard is automatically removed from the list. All inspection data, alarm information, rectification records, review reports, and infrared thermal imaging images are completely stored in the data storage module and linked to the corresponding location of the distribution cabinet in the BIM model, forming a complete lifecycle management file for safety hazards.

[0065] The above specific embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A management system for violations and safety hazards at engineering construction sites, characterized in that, It includes a robot dog inspection terminal, an edge computing module, a cloud management platform, a mobile APP, and a data storage module; The robot dog inspection terminal integrates a multimodal perception module, an audible and visual alarm module, a positioning module, and a communication module. The robot dog inspection terminal collects environmental data and worker behavior data from the construction site, receives control commands from the cloud management platform, and performs inspection, alarm, and verification tasks. The edge computing module is mounted on the robot dog inspection terminal. The edge computing module has a built-in customized AI recognition model for engineering scenarios. The edge computing module processes the multi-source data collected by the robot dog inspection terminal in real time, completes the identification of violations and the investigation of safety hazards, generates identification results and preliminary alarm information, and uploads key data to the cloud management platform. The cloud management platform receives the identification results, alarm information and inspection data uploaded by the edge computing module. The cloud management platform classifies and manages violations and safety hazards, issues alarms at different levels, assigns tasks, tracks rectification and conducts data statistical analysis. At the same time, it links with the engineering BIM model and digital twin system for data linkage. The mobile app synchronizes data with the cloud management platform in real time. The mobile app receives alarm information and rectification tasks pushed by the cloud management platform, provides feedback on the rectification status, and initiates a review application. The data storage module stores the raw data collected by the robot dog inspection terminal, the recognition results generated by the edge computing module, the control data of the cloud management platform, and the closed-loop processing records of the entire process.

2. The construction site violation and safety hazard management system according to claim 1, characterized in that, The robot dog inspection terminal has adaptive gait adjustment capability, and its protection level is not lower than IP67 and it has explosion-proof capability. The positioning module adopts SLAM and BIM fusion navigation technology, combined with IMU inertial measurement unit, and the positioning error of the positioning module is not greater than 10cm.

3. The construction site violation and safety hazard management system according to claim 2, characterized in that, The multimodal sensing module includes a 4K low-light industrial camera, an infrared thermal imaging probe, a millimeter-wave sensor, and an environmental sensor. The 4K low-light industrial camera captures video streams of worker behavior and images of equipment status at the construction site; The infrared thermal imaging probe captures hidden heat sources at the construction site. The millimeter-wave sensor detects the distance between the workers and the hazardous area; The environmental sensors collect data on temperature, humidity, dust concentration, and toxic gas concentration at the construction site.

4. The construction site violation and safety hazard management system according to claim 1, characterized in that, The edge computing module is equipped with an AI inference chip; The edge computing module preprocesses the data collected by the multimodal perception module, and the preprocessing includes adaptive histogram equalization and median filtering. The edge computing module uses online distillation technology to migrate cloud-based model knowledge to a lightweight model at the edge.

5. The construction site violation and safety hazard management system according to claim 4, characterized in that, The customized AI recognition model for engineering scenarios is based on deep learning algorithms and trained using labeled data from the construction site. The model can identify behaviors such as workers not wearing safety helmets, smoking in violation of regulations, entering high-risk work areas in violation of regulations, not wearing safety ropes while working at heights, and operating electrical equipment in violation of regulations, as well as the conditions of electrical equipment overheating, excessive toxic gases, and loose rocks.

6. The construction site violation and safety hazard management system according to claim 1, characterized in that, The cloud-based management platform includes a closed-loop hazard management module. The hidden danger closed-loop management module performs secondary verification on the received preliminary alarm information, automatically dispatches rectification tasks to the mobile APP of the corresponding management personnel according to the alarm risk level, controls the robot dog inspection terminal to go to the location of the hidden danger to perform the review task, and automatically cancels the number of hidden dangers that pass the review.

7. The construction site violation and safety hazard management system according to claim 1, characterized in that, The cloud management platform also includes an inspection management module; The inspection management module generates inspection paths based on the BIM model of the engineering site and electronic fences, supports adjusting the inspection frequency of different areas, remotely controls the inspection status of the robot dog inspection terminal, and obtains inspection progress, location information and real-time video stream in real time.

8. The construction site violation and safety hazard management system according to claim 1, characterized in that, The data storage module adopts a combination of cloud-based distributed storage and local caching. The data storage module encrypts the stored data, and the storage period of the data storage module is no less than the construction period plus 5 years; the data storage module has a data backup function.

9. The construction site violation and safety hazard management system according to claim 1, characterized in that, The communication module adopts 4G and 5G dual-mode communication, and the transmission delay of the communication module is no more than 50ms; The communication module uploads the identification results, alarm information and inspection data to the cloud management platform, and at the same time receives inspection path adjustment and task assignment control instructions issued by the cloud management platform.

10. A method for managing violations and safety hazards at engineering construction sites, characterized in that, The management system for violations and safety hazards at construction sites as described in any one of claims 1-9 includes: S1. Start the robot dog inspection terminal through the cloud management platform or mobile APP, select the preset or custom inspection path, and set the inspection frequency. S2. The robot dog inspection terminal moves autonomously along the inspection path. The positioning module obtains location information in real time and associates it with the BIM model. The multimodal perception module synchronously collects images, videos and environmental parameter data of the construction site and transmits them to the edge computing module. S3, the edge computing module preprocesses the collected data, uses a customized AI recognition model for engineering scenarios to identify violations and investigate safety hazards, and generates recognition results and preliminary alarm information; S4. The edge computing module uploads the recognition results, raw data and preliminary alarm information to the cloud management platform, and at the same time triggers the sound and light alarm of the robot dog inspection terminal. S5, the cloud management platform performs secondary verification of alarm information and assigns rectification tasks to the corresponding managers' mobile apps according to the alarm risk level; S6. Managers receive rectification tasks via mobile app, go to the site to handle them and report the rectification progress, and upload rectification materials before initiating a review application. S7, the cloud management platform controls the robot dog inspection terminal to go to the location of the hidden danger for secondary inspection, and verifies the rectification effect through the customized AI recognition model of the engineering scenario and generates a review report; S8 and the cloud management platform review and verify reports, automatically canceling the numbers of hazard reports that pass the verification, and the data storage module stores the entire process of handling records and associates them with the BIM model.

Images