Construction site quality management system and method based on ai large model

Abstract

This invention relates to the field of smart construction technology, and discloses a construction site quality management system and method based on an AI-powered large-scale model. It includes a data acquisition device, a multimodal data analysis and processing unit, a task and personnel association unit, a work scenario identification unit, a quality assessment unit, a risk warning unit, and a performance evaluation unit. The data acquisition device collects construction site videos, images, and personnel identity and location information; the multimodal data analysis and processing unit integrates and analyzes multi-source data and makes intelligent decisions; the task and personnel association unit binds personnel identities to construction tasks; the work scenario identification unit identifies the construction area and content; the quality assessment unit automatically determines the construction quality according to preset standards; the risk warning unit generates warnings based on the judgment results and pushes them to the responsible person; the performance evaluation unit automatically generates a performance report based on task completion status, quality judgment results, and warning responses. This invention achieves a closed-loop process for construction management, improving management efficiency and objectivity.

Description

Technical Field

[0001] This invention relates to the field of intelligent construction technology, specifically to a construction site quality management system and method based on an AI-powered large-scale model. Background Technology

[0002] As the construction industry transforms towards digitalization and intelligence, the demand for refined and real-time construction site management is increasing. Traditional construction site management mainly relies on manual inspections and paper records, which are gradually revealing limitations in efficiency and accuracy in areas such as personnel management, quality supervision, and task coordination.

[0003] Chinese patent CN113283753A discloses a construction site personnel safety management system. This system includes a personnel dynamic monitoring unit, a personnel real-name registration unit, an AI intelligent recognition unit, a three-level safety education unit, a VR safety education unit, a VR quality model unit, a BIM model unit, a participant personnel data analysis unit, an attendance management unit, a central processing unit, and a cloud platform. The solution provided in this application is applicable to the strict management of factors such as the work positions, schedules, access permissions to and from the construction site, personnel distribution, and the flow of safety materials for workers. It achieves real-time and accurate positioning of workers and equipment, establishing a complete and real-time management information system to implement responsibilities, improve the technical level of safe production, and ensure safe production. In particular, it can accurately and quickly identify the specific location and position of people in distress during disasters, improving rescue efficiency and effectiveness.

[0004] The aforementioned patents generally have the following shortcomings in construction site management: First, the personnel management function is relatively independent and cannot dynamically link personnel identity information with their specific construction tasks and work areas, resulting in limited monitoring of personnel on-duty status and work performance; second, there is a lack of automated performance evaluation methods based on objective data, making it impossible to quantitatively evaluate the quality and timeliness of work completion by management and construction personnel. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a construction site quality management system and method based on an AI large model, which can be used to realize intelligent closed-loop management of construction sites.

[0006] The technical solution adopted by this invention to solve its technical problem is a construction site quality management system based on an AI large-scale model, including a data acquisition device for collecting video, images, personnel identity and location information of the construction site, and also including:

[0007] A multimodal data analysis and processing unit, communicatively connected to the data acquisition device, is used to fuse and analyze video image data, personnel identity data, and location data, and to make intelligent decisions.

[0008] The task and personnel association unit is communicatively connected to the multimodal data analysis and processing unit, and is used to dynamically bind personnel identity information with construction tasks;

[0009] The work scene recognition unit is communicatively connected to the multimodal data analysis and processing unit, and is used to automatically identify the construction area and construction content based on video image data;

[0010] The quality assessment unit is communicatively connected to the multimodal data analysis and processing unit and the work scenario identification unit, and is used to automatically determine the construction quality according to preset standards.

[0011] The risk warning unit, which is communicatively connected to the quality assessment unit, is used to generate warning information based on the quality assessment results and push it to the responsible person.

[0012] The performance evaluation unit is communicatively connected to the task and personnel association unit, the quality assessment unit, and the risk warning unit, and is used to automatically generate personnel performance reports based on task completion status, quality judgment results, and warning response status.

[0013] Furthermore, the data acquisition device includes a high-definition network camera, a facial recognition access control machine, an IoT positioning tag, a drone, and a robot dog.

[0014] Furthermore, the multimodal data analysis and processing unit employs a deep learning model based on Transformer and convolutional neural networks.

[0015] Furthermore, it also includes an edge computing unit, which is disposed between the data acquisition device and the multimodal data analysis and processing unit, and is used to preprocess the acquired data.

[0016] Furthermore, the work scene recognition unit automatically identifies the project area, floor, and type of construction activity by comparing the video footage with the pre-loaded BIM model.

[0017] Furthermore, the quality assessment unit automatically compares and judges the construction operations or results collected in real time according to the preset digital construction acceptance standards.

[0018] Furthermore, the risk warning unit categorizes warning information into four levels—red, orange, yellow, and blue—based on risk level, and pushes these warnings through mobile applications, instant messaging tools, or on-site broadcasting systems.

[0019] Furthermore, the risk warning unit is also used to track the rectification progress corresponding to the warning and to close the warning after the rectification is completed.

[0020] Furthermore, the performance evaluation unit automatically generates performance reports based on the timeliness of task completion, quality pass rate, and early warning response efficiency of personnel, and links them to the salary system.

[0021] The construction site management method based on AI large model is characterized by the following steps:

[0022] S1. Personnel Real-Name Registration and Intelligent Division of Labor: Enter personnel identity information and bind personnel identity information with construction tasks;

[0023] S2. Automatic Recognition of Construction Area and Content: Collects video image data from the construction site, analyzes the video image data using an AI model, and identifies the construction area and construction content.

[0024] S3. Intelligent Judgment of Quality and Management Behavior: The construction process is analyzed through an AI model, and the analysis results are compared with preset standards to generate quality judgment results;

[0025] S4. Early Warning Generation and Push: Generate early warning information based on the quality judgment results and push the early warning information to the responsible person;

[0026] S5. Rectification Tracking and Closed-Loop Management: Record the rectification process of the responsible person and conduct acceptance testing after rectification is completed;

[0027] S6. Performance Evaluation: Automatically generate personnel performance reports based on task completion status, quality judgment results, and early warning response status.

[0028] The beneficial effects of this invention are: 1. By using AI for automatic identification and judgment, management personnel are freed from tedious on-site inspections and paper records. The efficiency of quality inspection is significantly improved compared to the traditional manual mode. The response time for hidden dangers is shortened from hours to minutes or even seconds. The average daily management area per person can be increased from the traditional 500 square meters to 2000 square meters. The quality acceptance time is shortened by 60%, which greatly improves the overall efficiency of construction management.

[0029] 2. Based on AI and big data analysis and judgment, the interference of human subjective factors is minimized. Quality judgment is automatically compared according to digital standards, and performance evaluation is automatically calculated based on objectively recorded data, making quality judgment and performance evaluation more objective, consistent and credible, and avoiding the bias, omissions or personal factors that may exist in manual inspection and evaluation.

[0030] 3. The system realizes a closed-loop digital management of the entire process from task allocation, process supervision, problem discovery, early warning push to rectification feedback. Every problem discovered will generate an early warning work order, which will be pushed to the responsible person, track the rectification progress, and accept the rectification results to ensure that the problem is effectively resolved. This forms a complete closed loop of "discovery-rectification-acceptance-recording" to avoid problems being missed or delayed.

[0031] 4. Through real-time monitoring and proactive early warning, safety accidents and quality defects can be effectively prevented, and rework rates can be reduced. Projects using this system can achieve a significant reduction in the accident rate, reduce rework due to quality defects and save costs, while also reducing project delays and economic losses caused by quality issues, thus improving the overall efficiency of the project.

[0032] 5. The system generates multi-dimensional data analysis reports, such as regional risk heat maps, job risk profiles, individual performance reports, and quality trend analyses, which provide precise data support for project managers to optimize resource allocation, adjust construction plans, strengthen safety training, and improve management measures. This transforms management decisions from experience-driven to data-driven, improving the scientific nature and effectiveness of decision-making. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0034] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0035] like Figure 1As shown, this invention relates to an AI-based large-scale model-based construction site quality management system for achieving intelligent closed-loop management of construction sites. The system includes: a data acquisition device for collecting video, images, personnel identity, and location information from the construction site; a multimodal data analysis and processing unit, communicatively connected to the data acquisition device, for fusing and analyzing video image data, personnel identity data, and location data, and making intelligent decisions; a task and personnel association unit, communicatively connected to the multimodal data analysis and processing unit, for dynamically binding personnel identity information with construction tasks; a work scene identification unit, communicatively connected to the multimodal data analysis and processing unit, for automatically identifying construction areas and content based on video image data; a quality assessment unit, communicatively connected to the multimodal data analysis and processing unit and the work scene identification unit, for automatically judging construction quality according to preset standards; a risk warning unit, communicatively connected to the quality assessment unit, for generating warning information based on quality judgment results and pushing it to the responsible person; and a performance evaluation unit, communicatively connected to the task and personnel association unit, the quality assessment unit, and the risk warning unit, for automatically generating personnel performance reports based on task completion status, quality judgment results, and warning response status.

[0036] The multimodal data analysis and processing unit of this invention, as the core of the system, can integrate and understand various types of data and make intelligent decisions, achieving collaborative perception of all elements of the construction site. The task and personnel association unit dynamically binds personnel identity information with specific construction tasks, solving the problem of real-name management becoming a mere formality. The work scene recognition unit automatically identifies construction areas and content through video analysis, providing spatial and activity context information for quality assessment. The quality assessment unit automatically makes judgments based on preset standards, overcoming the subjectivity and inefficiency of manual inspection. The risk warning unit realizes automatic problem detection and notification, and the performance evaluation unit automatically generates performance reports based on objective data, thus forming a complete management closed loop.

[0037] Furthermore, the data acquisition devices include high-definition network cameras, facial recognition access control machines, IoT positioning tags, drones, and robotic dogs. These diverse data acquisition devices can collect construction site information comprehensively and without blind spots. High-definition network cameras provide fixed-point video surveillance, facial recognition access control machines verify personnel entry and exit, IoT positioning tags track personnel location in real time, and drones and robotic dogs can flexibly inspect hard-to-reach areas, thus ensuring the integrity and coverage of data collection.

[0038] Furthermore, the multimodal data analysis and processing unit employs a deep learning model based on Transformer and convolutional neural networks. The Transformer architecture possesses powerful sequence data processing capabilities and an attention mechanism, enabling it to capture temporal relationships and key information in video streams. Convolutional neural networks excel at image feature extraction and spatial pattern recognition. The combination of the two enables a deep fusion and understanding of multimodal data, significantly improving the accuracy and robustness of recognition.

[0039] Furthermore, the system also includes an edge computing unit, which is located between the data acquisition device and the multimodal data analysis and processing unit to preprocess the acquired data. The edge computing unit performs preliminary processing at the network edge, close to the data source, reducing cloud bandwidth pressure, minimizing data transmission latency, and achieving second-level response times. This makes it particularly suitable for early warning scenarios with high real-time requirements.

[0040] Furthermore, the work scene recognition unit automatically identifies the project area, floor, and type of construction activity by comparing video footage with a pre-loaded BIM model. The BIM model contains complete three-dimensional geometric information and component attributes of the building. By spatially registering and semantically matching real-time video footage with the BIM model, the system can accurately locate the specific project location corresponding to the camera's viewpoint and determine the type of ongoing construction activity using image recognition technology. This provides precise spatial context for subsequent quality assessment and personnel management.

[0041] Furthermore, the quality assessment unit automatically compares and judges the real-time collected construction operations or results based on the preset digital construction acceptance standards. The digital construction acceptance standards transform traditional textual specifications into quantifiable and machine-recognizable parameter indicators, such as pipe slope range, support spacing threshold, and concrete surface flatness tolerance. The quality assessment unit extracts the geometric or process parameters of the construction results through computer vision technology and automatically compares them with the standards, thereby achieving objective and consistent quality judgment and avoiding the subjectivity and arbitrariness of manual inspection.

[0042] Furthermore, the risk warning unit categorizes warning information into four levels—red, orange, yellow, and blue—based on risk severity, and pushes these warnings through mobile applications, instant messaging tools, or on-site broadcasting systems. This four-level grading mechanism allows for differentiated handling based on the severity and urgency of the problem: red warnings correspond to major safety hazards or serious quality defects, requiring immediate work stoppage and rectification; orange warnings correspond to relatively serious issues, requiring priority handling; yellow warnings correspond to general issues, requiring timely rectification; and blue warnings correspond to minor deviations, which can be corrected within a reasonable timeframe. This multi-channel delivery method ensures that warning information is delivered to responsible personnel in a timely and accurate manner.

[0043] Furthermore, the risk warning unit is also used to track the rectification progress corresponding to the warning and close the warning after rectification is completed, forming a closed-loop management system. The system records key nodes such as the warning issuance time, the time the responsible person receives the warning, the rectification start time, the rectification completion time, and the acceptance time. It automatically calculates the response time and rectification time, and upgrades or reports warnings that are not processed within the time limit, ensuring that every problem is effectively resolved and preventing problems from being missed or delayed.

[0044] Furthermore, the performance evaluation unit automatically generates performance reports based on employees' timely task completion rate, quality pass rate, and early warning response efficiency, and links these reports to the compensation system. The system continuously records objective data such as each employee's task completion status, quality judgment results, and early warning response speed, periodically calculates various key performance indicators, and automatically generates individual performance scores and ratings. These results can be directly used as the basis for calculating performance-based pay and bonuses, achieving automation, transparency, and fairness in performance evaluation, and effectively motivating employees' work enthusiasm and sense of responsibility.

[0045] This invention also provides a construction management method based on the above system, comprising the following steps: S1, personnel real-name registration and intelligent division of labor: inputting personnel identity information and binding personnel identity information with construction tasks; S2, automatic identification of construction area and content: collecting video image data of the construction site, analyzing the video image data through an AI model and identifying the construction area and construction content; S3, intelligent judgment of quality and management behavior: analyzing the construction operation process through an AI model, comparing the analysis results with preset standards, and generating quality judgment results; S4, early warning generation and push: generating early warning information based on the quality judgment results and pushing the early warning information to the responsible person; S5, rectification tracking and closed-loop management: recording the rectification process of the responsible person and conducting acceptance after rectification is completed; S6, performance evaluation: automatically generating personnel performance reports based on task completion, quality judgment results, and early warning response.

[0046] Example 1

[0047] like Figure 1As shown, the AI-based construction site quality management system comprises data acquisition devices, a multimodal data analysis and processing unit, and an application unit. These units interact and transmit control commands via standardized interfaces. The data acquisition layer includes various data acquisition devices deployed at key locations throughout the construction site, consisting of high-definition network cameras, facial recognition access control machines, IoT positioning tags, drones, and robotic dogs, among other hardware. High-definition network cameras are deployed at main entrances, work surfaces, and high points, providing continuous video surveillance at fixed locations. The number of cameras is determined by the construction site area, with one camera deployed at a density of one per 500 to 1000 square meters to ensure a monitoring coverage rate of over 90%. Facial recognition access control machines are deployed at various entrances and exits. These machines communicate with the backend database in real-time via 4G or 5G networks to verify personnel identity and record entry and exit times. IoT positioning tags communicate with base stations via Bluetooth or LoRa protocols. Each person entering the construction site must wear a positioning tag, and the system tracks their location in real-time using a tag signal strength triangulation algorithm. Drones are used for regular inspections of areas inaccessible to fixed cameras, such as high-altitude areas, narrow passages, and dangerous zones; robot dogs are used for mobile inspections on the ground level, particularly suitable for areas such as basements, utility tunnels, and equipment rooms. Video, images, identity, and location data collected by each device are transmitted to the edge computing layer in real time.

[0048] The edge computing unit is responsible for real-time preprocessing and preliminary analysis of the collected data. Processing includes video frame extraction, image denoising and enhancement, target detection, facial feature extraction and comparison, location data denoising, and trajectory smoothing. The edge computing layer runs a lightweight deep learning model to perform preliminary identification and tracking of targets such as people, vehicles, and equipment in the video footage, outputting structured data. For facial images, feature extraction and preliminary comparison are performed; for location data, matching is performed to determine whether personnel are within their assigned work areas; for inspection data from drones and robotic dogs, real-time stitching and preliminary analysis are performed to identify obvious quality defects or safety hazards. The preprocessed data is compressed and uploaded to the cloud platform layer. Simultaneously, the edge layer can independently complete preliminary early warning tasks with high real-time requirements, achieving rapid response. The edge layer and the cloud platform layer exchange data and transmit control commands bidirectionally via a communication protocol.

[0049] The multimodal data analysis and processing unit adopts a deep learning model architecture based on Transformer and convolutional neural networks, and uses a large amount of construction site video, quality defect images, BIM models, and personnel work records for pre-training and fine-tuning. The unit receives multi-source data from data acquisition devices, as well as building component information loaded from the BIM database, task allocation information obtained from the project management system, and quality acceptance standards retrieved from the specification database. The processing flow includes three stages: data fusion, task inference, and result output. First, an attention mechanism is used to align features and fuse semantics of data from different modalities, generating a unified multimodal feature representation. Then, specialized inference heads are built for different business functions such as personnel-task matching, scene recognition, quality judgment, and risk assessment, outputting task-related decision results. Finally, the inference results are converted into structured data and output to various business microservices in the application layer through a standard API interface. This unit has continuous learning capabilities, enabling continuous model optimization to achieve high accuracy in personnel identification, scene recognition, and quality defect detection while maintaining a low false alarm rate.

[0050] The application unit mainly includes: The task and personnel association unit is responsible for personnel registration, identity verification, intelligent task allocation, and worker profiling. Project managers input basic personnel information, including name, ID number, contact information, job type, skill level, and professional qualification certificates, through a mobile app or web backend, and collect facial biometric data via facial recognition access control. The system can connect to authoritative data sources for identity verification to ensure information authenticity. Each time personnel enter the construction site, they undergo identity verification through facial recognition access control, and the system records the entry time and access control location. In the intelligent task allocation phase, project managers define a work breakdown structure through the system, decomposing the overall project objectives into several sub-tasks. Each sub-task includes attributes such as task name, work content, work area, planned time, and quality acceptance standards. Based on personnel's job type, skill level, and historical performance, the system automatically recommends suitable responsible persons for each sub-task. After project managers confirm, the task and personnel are bound together, and the responsible person is notified via push notification. Meanwhile, the system continuously records the work behavior data of each person, including attendance time, on-duty time, work area trajectory, number of tasks completed, quality pass rate, early warning response speed, etc., to generate multi-dimensional worker profiles to assist in subsequent division of labor and performance evaluation.

[0051] Work Scene Recognition Unit: Automatically identifies construction areas and construction content. The system spatially registers real-time video footage with a pre-loaded BIM model to determine the engineering area, floor, and specific component location within the footage. The registration process includes feature point extraction, feature matching, and pose estimation, achieving precise correspondence between video footage and the BIM model through computer vision technology. A construction activity classification model is then used to identify the type of ongoing construction activity, such as earthwork excavation, rebar tying, formwork installation, concrete pouring, and pipeline installation. The identification results are fused with the spatial registration results to generate structured scene description information, providing contextual information for quality assessment and risk evaluation.

[0052] Quality Assessment Unit: This unit automatically assesses construction quality based on digital construction acceptance standards. The system selects appropriate quality inspection algorithms according to the type of construction activity, extracting key geometric or process parameters from the video, such as rebar spacing, pipe slope, concrete surface defect dimensions, and support spacing. Parameter extraction employs deep learning-based semantic segmentation, object detection, and key point detection technologies. The extracted measured parameters are automatically compared with the digital construction acceptance standards retrieved from the specification database to determine if they are within the allowable error range, resulting in a judgment of "qualified," "unqualified," or requiring manual review. The judgment result includes the judgment conclusion, measured parameter values, standard parameter values, and deviations, and is output to the risk warning unit and stored in the quality database for subsequent traceability and statistical analysis.

[0053] The risk warning unit generates tiered warning information based on quality assessment results and pushes it to the responsible personnel. Warning levels are divided into multiple levels based on the severity of the problem: red, orange, yellow, and blue. A red warning corresponds to a major safety hazard or serious quality defect, requiring immediate work stoppage for rectification and reporting to the project director; an orange warning corresponds to a relatively serious problem, requiring rectification within 4 hours; a yellow warning corresponds to a general problem, requiring rectification within 24 hours; and a blue warning corresponds to minor deviations or suggestive information, recommending correction within a reasonable timeframe. Warning content includes the anomaly type, location coordinates (building, floor, room, or component number), on-site images (extracted from video streams and marked with defect areas), and rectification requirements (automatically generated rectification measure suggestions based on standards and specifications). Warning information is generated in structured data format and stored in the warning database, with each warning record assigned a unique number. The system determines the responsible personnel for warnings based on task binding relationships and pushes warning information through multiple channels such as mobile applications, instant messaging tools, and on-site broadcasting systems to ensure timely receipt by the responsible personnel. The system tracks the rectification progress of warning work orders, recording key time nodes such as warning generation, push, receipt, rectification initiation, rectification completion, and acceptance. After receiving the alert via the mobile app, the responsible person can click to start rectification and upload photos or videos after completion. The system automatically verifies the rectification results or transfers them to manual verification. Once verified, the alert is closed, forming a closed-loop management system. For alerts that are not processed within the specified time, the system automatically triggers an escalation reminder mechanism, sending a reminder to the responsible person's supervisor to ensure that the problem is resolved in a timely manner.

[0054] The performance evaluation unit is responsible for automatically generating employee performance reports based on objective data and linking them to the compensation system. The workflow of the performance evaluation unit includes KPI recording, performance calculation, report generation, and compensation linkage.

[0055] In the KPI recording process, work behavior data for each employee is continuously collected from multiple data sources, including the task and personnel association unit, quality assessment unit, and risk warning unit. The recorded key performance indicators include:

[0056] Attendance rate (actual attendance days / planned attendance days × 100%) is calculated from the entry and exit records of the facial recognition access control machine.

[0057] On-duty time (the actual time spent at the construction site each day) is calculated using location data from IoT positioning tags. Only when personnel spend more than 80% of their assigned work area is it counted as valid on-duty time.

[0058] The on-time completion rate (number of tasks completed on time / total number of tasks × 100%) is calculated using task status data from the task-personnel association unit. If the actual completion time of a task is later than the planned completion time, it is considered a delay.

[0059] The quality pass rate (number of jobs judged as qualified / total number of jobs × 100%) is calculated by statistically analyzing the quality judgment results of the quality assessment unit.

[0060] The early warning response efficiency (average early warning response time and average early warning rectification time) is calculated from the early warning work order data of the risk early warning unit. The early warning response time is defined as the difference between the time when the responsible person receives the early warning and the time when the early warning is pushed. The early warning rectification time is defined as the difference between the time when the early warning work order is accepted and the time when rectification is started.

[0061] The number of violations (such as not wearing a helmet, smoking, and other unsafe behaviors) is automatically detected and recorded by the behavior recognition model of the multimodal data analysis and processing unit.

[0062] The performance evaluation unit summarizes the above KPI records by personnel. Each person's KPI data is stored in the performance database in time series format, and the data is updated once a day to ensure the timeliness and accuracy of the performance data.

[0063] In the performance calculation step, the performance evaluation unit quantifies and scores the comprehensive performance of each employee according to a preset performance evaluation algorithm. The performance evaluation algorithm adopts a weighted summation model, with different weight coefficients assigned to each KPI indicator. The weight coefficients are configured by the project manager according to the project characteristics and management priorities. A typical weight configuration is: attendance rate weight 0.15, on-duty time weight 0.10, task completion timeliness weight 0.25, quality pass rate weight 0.30, and early warning response efficiency weight 0.2. The performance score calculation formula is: Performance Score = Attendance Rate × 0.15 + On-duty Time Score × 0.10 + Task Completion Timeliness × 0.25 + Quality Pass Rate × 0.30 + (100 - Average Early Warning Response Time / Minute) × 0.01 + (100 - Average Early Warning Rectification Time / Hour) × 0.01 - Number of Violations. The performance score is out of 100 points. A score below 60 is unqualified, 60 to 75 is qualified, 75 to 90 is good, and above 90 is excellent. The performance evaluation unit performs performance calculations on a monthly cycle. The performance calculation task is automatically triggered on the last day of each month to summarize and calculate the monthly performance data of all personnel and generate individual performance scores and ratings.

[0064] In the report generation step, the performance evaluation unit automatically generates individual performance reports and overall project performance reports based on the performance calculation results.

[0065] In the compensation linkage step, the performance evaluation unit automatically links individual performance scores with the company's compensation system. The compensation system typically adopts a "basic salary + performance-based salary" structure. The performance-based salary is calculated proportionally based on the performance score. A typical compensation calculation formula is: Monthly net salary = Basic salary + Performance-based salary base × Performance coefficient. The correspondence between the performance coefficient and the performance level is as follows: Unsatisfactory (performance score below 60) performance coefficient 0.5, Satisfactory (performance score 60 to 75) performance coefficient 1.0, Good (performance score 75 to 90) performance coefficient 1.2, Excellent (performance score above 90) performance coefficient 1.5. After the monthly performance calculation is completed, the performance evaluation unit automatically generates a compensation settlement list and pushes it to the financial system or human resources system through a standard interface. The financial system completes the salary payment based on the compensation settlement list, realizing the automated linkage between performance evaluation and compensation incentives. The performance evaluation unit also generates a summary compensation report for project finance personnel and the company's finance department to check and approve.

[0066] The embodiments described herein are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A construction site quality management system based on an AI large-scale model, including a data acquisition device for collecting video, images, personnel identification, and location information of the construction site, characterized in that... Also includes: A multimodal data analysis and processing unit, communicatively connected to the data acquisition device, is used to fuse and analyze video image data, personnel identity data, and location data, and to make intelligent decisions. The task and personnel association unit is communicatively connected to the multimodal data analysis and processing unit, and is used to dynamically bind personnel identity information with construction tasks; The work scene recognition unit is communicatively connected to the multimodal data analysis and processing unit, and is used to automatically identify the construction area and construction content based on video image data; The quality assessment unit is communicatively connected to the multimodal data analysis and processing unit and the work scenario identification unit, and is used to automatically determine the construction quality according to preset standards. The risk warning unit, which is communicatively connected to the quality assessment unit, is used to generate warning information based on the quality assessment results and push it to the responsible person. The performance evaluation unit is communicatively connected to the task and personnel association unit, the quality assessment unit, and the risk warning unit, and is used to automatically generate personnel performance reports based on task completion status, quality judgment results, and warning response status.

2. The construction site quality management system based on an AI large model as described in claim 1, characterized in that... The data acquisition device includes a high-definition network camera, a facial recognition access control machine, an IoT positioning tag, a drone, and a robot dog.

3. The construction site quality management system based on an AI large model according to claim 1, characterized in that... The multimodal data analysis and processing unit adopts a deep learning model based on Transformer and convolutional neural networks.

4. The construction site quality management system based on an AI large model according to claim 1, characterized in that... It also includes an edge computing unit, which is located between the data acquisition device and the multimodal data analysis and processing unit, and is used to preprocess the acquired data.

5. The construction site quality management system based on an AI large model according to claim 1, characterized in that... The work scene recognition unit automatically identifies the project area, floor, and type of construction activity by comparing the video footage with the pre-loaded BIM model.

6. The construction site quality management system based on an AI large model according to claim 1, characterized in that... The quality assessment unit automatically compares and judges the construction operations or results collected in real time according to the preset digital construction acceptance standards.

7. The construction site quality management system based on an AI large model according to claim 1, characterized in that... The risk warning unit divides the warning information into four levels: red, orange, yellow, and blue, according to the risk level, and pushes them through mobile applications, instant messaging tools, or on-site broadcasting systems.

8. The construction site quality management system based on an AI large model according to claim 7, characterized in that... The risk warning unit is also used to track the rectification progress corresponding to the warning and to close the warning after the rectification is completed.

9. The construction site quality management system based on an AI large model according to claim 1, characterized in that... The performance evaluation unit automatically generates performance reports based on the timeliness of task completion, quality pass rate, and early warning response efficiency of personnel, and links them to the salary system.

10. A construction site quality management method based on an AI large-scale model, characterized in that... This includes the following steps: S1. Personnel Real-Name Registration and Intelligent Division of Labor: Enter personnel identity information and bind personnel identity information with construction tasks; S2. Automatic Recognition of Construction Area and Content: Collects video image data from the construction site, analyzes the video image data using an AI model, and identifies the construction area and construction content. S3. Intelligent Judgment of Quality and Management Behavior: The construction process is analyzed through an AI model, and the analysis results are compared with preset standards to generate quality judgment results; S4. Early Warning Generation and Push: Generate early warning information based on the quality judgment results and push the early warning information to the responsible person; S5. Rectification Tracking and Closed-Loop Management: Record the rectification process of the responsible person and conduct acceptance testing after rectification is completed; S6. Performance Evaluation: Automatically generate personnel performance reports based on task completion status, quality judgment results, and early warning response status.

Images