A work order closed-loop management system based on PDCA-SDCA double cycle

Abstract

The application relates to the technical field of work order management, in particular to a work order closed-loop management system based on PDCA-SDCA double circulation, which comprises a data acquisition module, a hidden danger analysis unit, a rule reasoning module, a work order distribution module, a verification feedback module and a knowledge optimization module. Through the structured feature deconstruction algorithm of the 5M1E, i.e. man (Man), machine (Machine), material (Material), method (Method), environment (Environment) and measurement (Measurement), and the safety rule knowledge base of graph structure storage, accurate identification and classification of hidden dangers are realized; the unmanned aerial vehicle rephotographing and image comparison technology is used to complete automatic verification of rectification effect; and the safety rule knowledge base is dynamically optimized through an incremental learning algorithm, for example, an online random forest algorithm is used to realize dynamic updating of the knowledge base. The application can improve the comprehensiveness and accuracy of hidden danger identification, ensure dynamic adjustment and closed-loop management of rectification task priority, and effectively improve the efficiency and reliability of safety management of a railway engineering construction site.

Description

Technical Field

[0001] This invention relates to the field of railway engineering construction site safety management technology, and more specifically, to a work order closed-loop management system based on the PDCA-SDCA dual-cycle. Background Technology

[0002] The ever-expanding scale of railway construction projects presents increasingly complex challenges to safety management at construction sites. Traditional methods for identifying safety hazards rely heavily on manual inspections and experience-based judgment, which are inefficient and have limited coverage. They are also limited in their ability to comprehensively manage multi-dimensional hazards across all railway engineering disciplines, including roadbeds, bridges, tunnels, tracks, station buildings, and electrical systems. Furthermore, the experience-based identification method is highly subjective and prone to overlooking critical hazards, affecting the accuracy and timeliness of safety management.

[0003] Currently, existing technologies for hazard management suffer from a disconnect in the closed-loop process. Hazard identification, rectification implementation, feedback verification, and the accumulation of experience and knowledge are typically scattered across different stages, lacking a unified information management platform. This makes it difficult to achieve real-time tracking and closed-loop management of the entire hazard handling process. This fragmentation not only reduces rectification efficiency but also hinders the continuous accumulation and optimization of hazard knowledge, limiting the continuous improvement capabilities of the safety management system. Furthermore, existing hazard identification methods are often one-dimensional, analyzing only certain construction elements or specific project types, lacking systematic and multi-dimensional risk identification methods. Especially in railway engineering, the complex construction environment involving multiple disciplines and stages places higher demands on the identification and early warning of safety hazards. Traditional methods have certain shortcomings in meeting the needs of modern and intelligent safety management.

[0004] Therefore, there is an urgent need to invent a work order management technology to address the shortcomings of existing railway engineering safety hazard management methods, such as disconnected closed-loop management, single identification dimensions, insufficient informatization and intelligence levels, and difficulty in achieving real-time tracking, accurate early warning, and continuous knowledge optimization throughout the entire process. Summary of the Invention

[0005] In view of this, the present invention proposes a work order closed-loop management system based on PDCA-SDCA dual cycle, which aims to solve the problems of existing railway engineering safety hazard management methods, such as disconnected closed-loop management, single identification dimensions, insufficient informatization and intelligence levels, and difficulty in achieving real-time tracking, accurate early warning and continuous knowledge optimization throughout the entire process.

[0006] This invention proposes a work order closed-loop management system based on the PDCA-SDCA dual-cycle model, comprising: The data acquisition module is configured to acquire construction site information based on drone inspection equipment, IoT sensors, and video surveillance devices. The hazard analysis unit communicates with the data acquisition module. The hazard analysis unit is equipped with several analysis channels, and each analysis channel corresponds to a different construction specialty. Based on the 5M1E feature deconstruction algorithm, the hazard is analyzed in a structured manner from multiple dimensions to generate standardized feature vectors. The rule reasoning module communicates with the hazard analysis unit. The rule reasoning module is configured to perform real-time scenario-based logical reasoning based on a safety rule knowledge base stored in a graph structure, determine the hazard classification, and generate rectification instructions. The work order dispatch module communicates with the rule reasoning module. The work order dispatch module is configured to receive rectification instructions and assign rectification tasks to the responsible person's terminal. The verification feedback module is connected to the data acquisition module and the responsible person's terminal. The verification feedback module is configured to automatically compare and analyze the rectification effect based on the image data uploaded by the drone or the responsible person's terminal, according to the CNN network. The knowledge optimization module is configured to update rule entries in the security rule knowledge base based on incremental learning algorithms and verification results, and trigger a new round of PDCA process.

[0007] Furthermore, the data acquisition module is equipped with a data buffer, which is divided into a fixed buffer and a temporary buffer. A fixed buffer is used to store data acquired at high frequencies, while a temporary buffer is used to store data acquired at low frequencies. When the fixed cache reaches the preset capacity, the temporary cache triggers a data migration command to migrate some data to the temporary cache, where the preset capacity is 1TB. When the timestamp of the migrated data reaches a preset time threshold, the temporary cache triggers a data cleanup command to remove the data that exceeds the time threshold, which is 7 days.

[0008] Furthermore, the hazard analysis unit responds to the data cleaning command and recalculates the feature weight coefficients based on the remaining data; When the change in the feature weight coefficient exceeds the preset threshold, the hidden danger analysis unit triggers a rematching instruction from the rule reasoning module.

[0009] Furthermore, the hazard analysis unit includes: The feature extraction submodule is connected to the data acquisition module and is configured to extract key attributes from the raw data. The feature mapping submodule is connected to the feature extraction submodule. The feature mapping submodule is configured to map the extracted key attributes to a predefined feature space to form a standardized feature vector. The feature fusion submodule communicates with the feature mapping submodule. The feature fusion submodule is configured to perform weighted fusion of feature vectors from multiple dimensions to generate a comprehensive feature representation. The weight coefficients are dynamically adjusted based on historical data statistical patterns.

[0010] Furthermore, the rule-based reasoning module includes: The rule matching submodule is configured as a security rule knowledge base stored based on a graph structure. The rule matching submodule is also configured to search for rule entries that match the current hazard feature vector based on a depth-first search algorithm. The risk assessment submodule is configured to calculate the risk level of potential hazards based on the matching results and generate corresponding early warning signals; The instruction generation submodule is configured to generate rectification instructions based on risk level and early warning signals, and then transmit the instructions to the work order dispatch module.

[0011] Furthermore, the work order dispatch module includes: The task queue management submodule is configured to prioritize rectification tasks according to the risk level and occurrence time of the hazards; The task assignment submodule is configured to select the optimal responsible person based on their responsibilities and geographical location information, and then push the rectification task to the responsible person's terminal. The task status tracking submodule is configured to record the execution progress of rectification tasks in real time and synchronize the status information to the verification feedback module. The responsible person's terminal is a smart handheld device running the Android system, which supports work order reception, image uploading, and progress feedback.

[0012] Furthermore, the verification feedback module includes: The image acquisition submodule is configured to acquire the rectified image data through drone re-shooting or uploading via the responsible person's terminal; The image alignment submodule is configured to extract image features based on a pre-trained ResNet-50 model. The image alignment submodule is also configured to comprehensively determine the rectification effect using the structural similarity index (SSIM) and mean squared error (MSE). The result determination submodule is configured to determine whether the rectification is qualified based on the comparison results and pass the determination results to the knowledge optimization module.

[0013] Furthermore, the knowledge optimization module includes: The rule update submodule is configured to add, delete, and modify rule entries in the security rule knowledge base based on the results of the verification feedback module. The rule evaluation submodule is configured to evaluate the validity of rule entries using a cross-validation algorithm and remove redundant rules. The rule publishing submodule is configured to publish updated rule entries to the rule reasoning module, triggering a new round of PDCA process.

[0014] Furthermore, the rule reasoning module responds to the rematching instruction and recalculates the risk level of the hazard based on the updated feature weight coefficients; When the risk level changes, the rule reasoning module triggers the task priority adjustment instruction from the work order assignment module.

[0015] Furthermore, the temporary buffer responds to data migration instructions and marks the timestamp of the migrated data as a reference point; When the reference point reaches a preset time threshold, the temporary cache triggers a data cleanup command to remove data that exceeds the time threshold in order to free up storage space.

[0016] Compared with existing technologies, the advantages of this invention are as follows: The data acquisition module enables real-time acquisition of multimodal information from the construction site, including multi-source information such as drone inspection images, IoT sensor data, and video surveillance streams, ensuring a comprehensive, dynamic, and accurate data foundation for hazard identification. After preprocessing, the acquired data is input to the hazard analysis unit. This unit, based on the 5M1E feature deconstruction algorithm, performs structured analysis and standardized characterization of potential hazards from six dimensions: "human, machine, material, method, environment, and measurement," thereby achieving a unified description and high-precision identification of safety hazards across different construction specialties. Building upon hazard identification, the rule-based reasoning module constructs multi-hop logical relationships using a graph-structured knowledge base, achieving automatic hazard classification and rectification instruction generation based on scenario-based logical reasoning. This module utilizes the scalability and associative storage characteristics of graph databases to organize and reason about hazard characteristics, environmental variables, and rectification strategies in a knowledge graph format, significantly reducing reliance on manual judgment and experience, and improving the intelligence and real-time nature of safety decision-making. To ensure the efficient implementation of rectification measures, the system includes a work order assignment module. This module automatically distributes rectification tasks to responsible personnel based on reasoning results, ensuring clear task responsibilities and transparency in the execution process, thus guaranteeing a rapid closed-loop rectification workflow. After rectification, the verification feedback module automatically compares and analyzes the before-and-after scenes using drone-retrieved images or photos uploaded by responsible personnel. The system employs a composite verification model combining CNN feature extraction and SSIM (Structural Similarity) algorithms to achieve unmanned and accurate evaluation of rectification effectiveness. When the detection results fail to meet standards, the system automatically upgrades the rectification level and re-initiates the closed-loop process, ensuring traceability and controllability throughout the entire hazard handling process. Furthermore, the knowledge optimization module dynamically updates the safety rule knowledge base based on verification feedback results. Through online learning and version control mechanisms, it supports adaptive optimization and rollback management of rules, triggering a new PDCA cycle. This mechanism enables continuous knowledge accumulation and model self-evolution, giving the system continuous learning and self-correction capabilities, forming a dual-cycle driving model of "PDCA cycle driving hazard management optimization and SDCA cycle solidifying standardized results." Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A functional block diagram of a work order closed-loop management system based on PDCA-SDCA dual-cycle provided in this embodiment of the invention; Figure 2 A flowchart illustrating a work order closed-loop management system based on a PDCA-SDCA dual-cycle model is provided for an embodiment of the present invention. Figure 3 This is a schematic diagram of the 5M1E feature deconstruction and graph rule matching process provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the PDCA-SDCA dual-cycle collaborative working mechanism provided in an embodiment of the present invention; Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] like Figures 1-4 As shown in some embodiments of this application, this embodiment provides a work order closed-loop management system based on PDCA-SDCA dual-cycle, including: a data acquisition module, a hidden danger analysis unit, a rule reasoning module, a work order assignment module, a verification feedback module, and a knowledge optimization module.

[0020] Specifically, the data acquisition module is configured to acquire construction site information based on drone inspection equipment, IoT sensors, and video surveillance devices.

[0021] Specifically, the data acquisition module has a data cache area, which is divided into a fixed cache area and a temporary cache area. The fixed cache area is used to store high-frequency acquired data, and the temporary cache area is used to store low-frequency acquired data. When the fixed cache area reaches the preset capacity, the temporary cache area triggers a data migration command to migrate some data to the temporary cache area. When the timestamp of the migrated data reaches the preset time threshold, the temporary cache area triggers a data cleanup command to remove data that exceeds the time threshold.

[0022] Specifically, the temporary cache responds to data migration instructions and marks the timestamps of the migrated data as reference points. When the reference point reaches a preset time threshold, the temporary cache triggers a data cleanup instruction to remove data that exceeds the time threshold in order to free up storage space.

[0023] Understandably, the data acquisition module efficiently and in real-time collects multimodal information from the construction site, encompassing drone inspection equipment, IoT sensors, and video surveillance devices, thereby comprehensively acquiring multi-dimensional information on the construction environment, equipment operating status, personnel activities, and construction procedures. Through continuous acquisition and real-time caching of on-site data, the system can provide a reliable data foundation for subsequent hazard identification, analysis, and rectification. To improve data storage and processing efficiency, the data acquisition module employs a hierarchical caching management mechanism, including a fixed cache area and a temporary cache area. The fixed cache area primarily stores frequently acquired data, such as images frequently captured by drone inspections and real-time measurement data from key sensors, ensuring the system maintains rapid response and real-time processing capabilities even under high data acquisition frequencies. The temporary cache area stores low-frequency acquired data, such as periodic sensor records or non-critical video clips, preventing the fixed cache area from being occupied by low-frequency data and affecting system operating efficiency. When the data volume in the fixed cache area reaches a preset capacity, the system automatically triggers a data migration command, migrating some data to the temporary cache area. This migration process is based on data timestamps, using the data acquisition time as a reference point to achieve precise management of the data lifecycle. The migration mechanism not only alleviates the storage pressure on the fixed cache area but also maintains the real-time nature of high-frequency data, enabling the system to maintain stable performance during continuous operation and providing a complete data chain for subsequent data analysis and hazard identification. The temporary cache area further undertakes the function of dynamic data maintenance. When the timestamp of migrated data reaches a preset time threshold, the system automatically triggers a data cleanup command to remove data exceeding the time threshold, thereby freeing up storage space and preventing long-term occupation of system storage resources. Through the timestamp reference point and threshold cleanup mechanism, the system can achieve automated management of cached data, ensuring that data storage is always in an optimal state while avoiding manual intervention, thus improving the system's intelligence and maintainability. Finally, this hierarchical cache management mechanism also has good scalability and adaptability. For different types and frequencies of data collection needs at the construction site, the system can flexibly adjust the capacity ratio of the fixed cache area and the temporary cache area, as well as the data cleanup strategy, to prioritize high-frequency critical data while also considering the storage and analysis needs of low-frequency data. This design not only improves data processing efficiency and system response speed but also provides a solid foundation for subsequent data-driven hazard analysis, rule reasoning, and closed-loop management.

[0024] It can be seen that by organically combining multimodal data acquisition, hierarchical cache management, data migration and automatic cleaning mechanisms, efficient acquisition, dynamic management and long-term availability maintenance of construction site data have been achieved, improving the system's data processing efficiency, stability and intelligence level, and providing reliable data support for the identification and closed-loop management of safety hazards in railway engineering.

[0025] Specifically, the hazard analysis unit is connected to the data acquisition module. The hazard analysis unit is equipped with several analysis channels, and each analysis channel corresponds to a different construction specialty. Based on the 5M1E feature deconstruction algorithm, the hazard is analyzed in a structured manner from six dimensions: people, machine, material, method, environment, and measurement, and a standardized feature vector is generated.

[0026] Specifically, the hazard analysis unit responds to the data cleaning command and recalculates the feature weight coefficients based on the remaining data; when the change in the feature weight coefficients exceeds a preset threshold, the hazard analysis unit triggers a rematching command from the rule reasoning module.

[0027] Specifically, the hazard analysis unit is based on the Neo4j graph database and uses the Cypher query language to implement depth-first search. Specifically, the hazard analysis unit includes: a feature extraction submodule, which communicates with the data acquisition module and is configured to extract key attributes from the raw data; a feature mapping submodule, which communicates with the feature extraction submodule and is configured to map the extracted key attributes to a predefined feature space to form standardized feature vectors; and a feature fusion submodule, which communicates with the feature mapping submodule and is configured to perform weighted fusion of feature vectors from multiple dimensions to generate a comprehensive feature representation; wherein the weight coefficients are dynamically adjusted according to historical data statistical patterns.

[0028] Understandably, a structured representation and intelligent response to construction site hazards are achieved through a multi-dimensional feature deconstruction and graph database-driven dynamic reasoning linkage mechanism. The core idea is to decompose complex safety hazard information into computable and associative feature vectors, and trigger system-level reasoning updates when data changes or feature weights become abnormal, thus forming an adaptive hazard identification and reasoning framework. First, the hazard analysis unit communicates with the data acquisition module, receiving real-time information from drone inspections, IoT sensors, and video surveillance through multi-source data interfaces. Multiple analysis channels are set up within the unit, each corresponding to different construction specialties (such as civil engineering, electrical, and track engineering) to ensure professional hazard analysis and parallel processing capabilities. During the analysis process, based on the 5M1E feature deconstruction algorithm, hazards are element-based decomposed from six dimensions: Man, Machine, Material, Method, Environment, and Measurement. Each dimension's features undergo three levels of processing—extraction, mapping, and fusion—to gradually form a quantifiable hazard feature model. The system comprises several sub-modules: Feature Extraction Submodule extracts key attributes from raw multi-source data, such as personnel work status, equipment operating parameters, material batches, construction method records, environmental monitoring data, and measurement deviations; Feature Mapping Submodule maps these key attributes to a predefined feature space, generating standardized feature vectors to achieve a unified representation of different types of hazard data; and Feature Fusion Submodule weights and fuses feature vectors from six dimensions to form a comprehensive feature representation reflecting the overall state of the hazard. Weight coefficients are dynamically adjusted based on the statistical patterns of historical sample data, enabling the model to adaptively update according to changes in the construction site environment and risk characteristics. When data cleaning is performed or an abnormal feature update request is received, the hazard analysis unit recalculates the feature weight coefficients. If the weight change exceeds a preset threshold, the system triggers a re-matching instruction from the rule reasoning module, prompting a recalculation of the hazard relationship rules in the knowledge graph. This linkage mechanism ensures real-time consistency between the hazard identification logic and the on-site data status. At the underlying implementation level, hazard association modeling is performed based on the Neo4j graph database. A depth-first search (DFS) query is used with the Cypher query language to retrieve multi-hop relationships between hazard nodes, enabling rapid reasoning from feature vectors to hazard categories, risk levels, and rectification strategies. Leveraging the scalability and depth-first traversal capabilities of graph structures, the system can perform high-dimensional, multi-path logical correlation analysis of potential hazards in complex construction scenarios. The hazard analysis unit, through a technical chain of "5M1E multi-dimensional feature modeling—dynamic feature fusion—graph database-linked reasoning," achieves standardized expression of hazard information, adaptive weight updates, and automatic rule reconstruction. This constructs a core hazard cognition system with dynamic perception, intelligent analysis, and self-learning capabilities, providing a solid data and knowledge foundation for subsequent reasoning, decision-making, and closed-loop management.

[0029] Specifically, the hazard analysis unit constructs a knowledge association structure based on the Neo4j graph database. Its technical principle lies in using the node-edge relationship model of the graph database to perform directed graph modeling of the characteristics of hazards at the construction site and their interdependencies. Graph query algorithms are used to achieve multi-hop association retrieval and deep logical reasoning of hazards. Specifically, the Neo4j graph database adopts an attribute graph model, abstracting hazard elements (such as personnel behavior, equipment status, material anomalies, construction method deviations, environmental factors, and measurement errors) as nodes, representing the causal, constraint, or dependency relationships between them as edges, and recording characteristic parameters such as weights, timestamps, and risk levels through properties. Compared with traditional relational databases, this structure has inherently high correlation and high query efficiency, making it particularly suitable for modeling and analyzing multi-factor interactions in complex safety hazard networks. After the graph database structure is constructed, the hazard analysis unit uses the Cypher query language to traverse and reason about the graph data. The Cypher language has semantic expression capabilities for pattern matching and path retrieval, and can describe complex hazard relationship query logic in a concise graph model. For example, by matching the pattern of "hazard node - causal path - result node," key factors or potential propagation paths that cause high-risk hazards can be quickly located. During hazard analysis, the system employs a depth-first search (DFS) strategy to traverse the graph structure at multiple levels. Its technical principle is to start from the target hazard node and explore downwards along the associated edges until the maximum depth is reached or a termination condition is met, thereby identifying potential multi-hop risk propagation chains. Compared to breadth-first search, depth-first search can discover deeper, indirect hazard relationships more quickly, helping to reveal hidden risk coupling paths and complex hazard sources. Furthermore, by combining node attribute weight calculation and path scoring mechanisms during DFS traversal, the system can prioritize search results based on risk, thus supporting intelligent decision-making by the hazard analysis unit in hazard identification, classification, and rule matching processes.

[0030] It can be seen that through the multi-dimensional feature extraction, mapping and fusion mechanism of the hidden danger analysis unit, the structured, standardized and dynamic calculation and processing of hidden dangers at the construction site are realized, supporting multi-professional and multi-dimensional risk analysis and automated early warning, and providing core intelligent processing capabilities for the PDCA-SDCA closed-loop safety management system.

[0031] Specifically, the rule reasoning module is connected to the hazard analysis unit. The rule reasoning module is configured to perform real-time scenario-based logical reasoning based on a safety rule knowledge base stored in a graph structure, determine the hazard classification, and generate rectification instructions.

[0032] Specifically, the rule reasoning module includes: a rule matching submodule, configured to store a safety rule knowledge base based on a graph structure, and further configured to use a depth-first search algorithm to find rule entries that match the current hazard feature vector; a risk assessment submodule, configured to calculate the risk level of the hazard based on the matching results and generate corresponding warning signals; and an instruction generation submodule, configured to generate rectification instructions based on the risk level and warning signals, and transmit the instructions to the work order dispatch module.

[0033] Specifically, the rule reasoning module responds to the rematching instruction and recalculates the risk level of the hidden danger based on the updated feature weight coefficients; when the risk level changes, the rule reasoning module triggers the task priority adjustment instruction of the work order assignment module.

[0034] Understandably, the rule reasoning module performs real-time scenario-based logical reasoning on the standardized feature vectors generated by the hazard analysis unit, thereby automatically determining the hazard classification and generating corresponding rectification instructions. The rule reasoning module communicates with the hazard analysis unit, receiving multi-dimensional feature vector inputs in real time and performing reasoning analysis based on a graph-structured safety rule knowledge base, achieving automation and intelligence in hazard identification and risk assessment. This module can efficiently handle complex, multi-dimensional construction hazards, ensuring the real-time nature and accuracy of risk management. Secondly, the rule reasoning module includes a rule matching submodule, a risk assessment submodule, and an instruction generation submodule: The rule matching submodule, based on a graph-structured safety rule knowledge base, uses a depth-first search algorithm to find rule entries that match the current hazard feature vector, achieving efficient mapping between multi-dimensional features and safety rules. This graph-structured matching method can take into account the logical relationships and dependency order between rules, improving rule retrieval speed and matching accuracy. The risk assessment submodule calculates the risk level of the hazard based on the matching results and generates corresponding warning signals. Through weighted analysis of different dimensional features and reference to historical data statistical patterns, this module can dynamically assess the potential hazards of the hazard, providing accurate risk classification and a reliable basis for generating rectification instructions. The instruction generation submodule generates rectification instructions based on risk levels and early warning signals, and transmits these instructions to the work order assignment module, enabling automated assignment and execution of hazard rectification tasks. This module can dynamically adjust the work order assignment order based on risk levels and construction site priorities, improving rectification efficiency and safety management response speed. The rule reasoning module also has dynamic adaptive capabilities. When it receives a re-matching instruction triggered by the hazard analysis unit, the rule reasoning module recalculates the hazard risk level based on the updated feature weight coefficients. If the risk level changes, the system automatically triggers a task priority adjustment instruction from the work order assignment module, ensuring that rectification resources and attention are prioritized for high-risk hazards. This dynamic adjustment mechanism enables the system to respond quickly to changes in the construction site environment and data updates, achieving closed-loop and real-time optimization of hazard management.

[0035] It can be seen that, through the multi-level structure and dynamic adaptive reasoning mechanism of the rule reasoning module, efficient matching of hidden danger characteristics and safety rules, dynamic assessment of risk levels, and automatic generation of rectification instructions are achieved, thus providing core technical support for the intelligent and safe management of railway engineering construction sites.

[0036] Specifically, the work order dispatch module communicates with the rule reasoning module and is configured to receive rectification instructions and assign rectification tasks to the responsible person's terminal.

[0037] Specifically, the work order assignment module includes: a task queue management submodule, configured to prioritize rectification tasks according to the risk level and occurrence time of the hazards; a task assignment submodule, configured to select the optimal responsible person based on the scope of responsibility and geographical location information of the responsible person, and push the rectification task to their terminal; and a task status tracking submodule, configured to record the execution progress of rectification tasks in real time and synchronize the status information to the verification feedback module.

[0038] Understandably, the work order dispatch module enables intelligent management and efficient allocation of rectification instructions generated by the rule-based reasoning module, ensuring that hazard rectification tasks are quickly assigned to specific responsible parties. The work order dispatch module communicates with the rule-based reasoning module, receiving rectification instructions in real time and automating the processing of hazard rectification tasks, thus improving the response speed and execution efficiency of on-site safety management. The work order dispatch module includes a task queue management submodule, a task dispatch submodule, and a task status tracking submodule: the task queue management submodule prioritizes rectification tasks based on hazard risk level and occurrence time. By prioritizing high-risk and newly discovered hazard tasks, this module ensures timely handling of critical hazards on the construction site, improving the targeting and effectiveness of safety management. The task dispatch submodule selects the optimal responsible party based on their responsibilities, skill level, and geographical location, pushing the rectification task to the responsible party's terminal. Through an intelligent personnel matching mechanism, the system ensures rapid response and efficient execution of tasks, while reducing manual scheduling costs and the possibility of human error in allocation. The task status tracking submodule records the execution progress of rectification tasks in real time and synchronizes the status information to the verification feedback module, providing data support for subsequent verification of rectification effects and closed-loop management. This module enables full traceability of the rectification process, making construction safety management more transparent and controllable.

[0039] It can be seen that through the task queue management, intelligent assignment and status tracking mechanism of the work order assignment module, the efficient allocation and full-process monitoring of hidden danger rectification tasks have been achieved, which has improved the timeliness, reliability and traceability of rectification execution, and provided key execution and feedback support for the PDCA-SDCA closed-loop management system.

[0040] Specifically, the verification feedback module is connected to the data acquisition module and the responsible person's terminal. The verification feedback module is configured to automatically compare and analyze the rectification effect based on the image data captured by the drone or uploaded by the responsible person's terminal.

[0041] Specifically, the verification feedback module includes: an image acquisition submodule, configured to acquire rectified image data through drone re-shooting or upload by the responsible person's terminal; an image comparison submodule, configured to extract image features based on a pre-trained ResNet-50 model, and the image comparison submodule is also configured to comprehensively judge the rectification effect using the structural similarity index (SSIM) and mean squared error (MSE); and a result judgment submodule, configured to determine whether the rectification is qualified based on the comparison results, and transmit the judgment result to the knowledge optimization module.

[0042] Specifically, the image alignment submodule uses the PyTorch framework, with ResNet-50 fine-tuned based on ImageNet pre-training.

[0043] Understandably, by utilizing deep convolutional neural network (ResNet-50) feature extraction combined with a structural similarity index (SSIM) and mean squared error (MSE) fusion judgment algorithm, automated image comparison and intelligent judgment of the rectification effect at construction sites can be achieved. This replaces the inefficient traditional method relying on manual review, enabling unmanned and quantifiable verification of hazard rectification. First, the verification feedback module communicates with the data acquisition module and the responsible person's terminal, acquiring on-site image data after rectification through the image acquisition submodule. This module supports two acquisition methods: one is to acquire panoramic images through drone re-shooting, achieving large-scale, high-viewpoint verification; the other is for the responsible person's terminal to upload locally captured images, enabling supplementary verification of detailed areas. This multi-channel acquisition mechanism ensures the authenticity and comprehensive coverage of the data source. After data acquisition, the image comparison submodule undertakes the core tasks of image feature extraction and difference assessment. This submodule is built on the PyTorch deep learning framework, using a ResNet-50 residual network as the backbone feature extractor. ResNet-50 was pre-trained on the ImageNet dataset and fine-tuned in this system using transfer learning to adapt to the specific feature distribution of construction site images. Its technical principle lies in alleviating the gradient vanishing problem of deep networks by introducing residual blocks, thereby achieving efficient extraction of deep features and enabling the model to accurately capture subtle structural changes, surface texture differences, and geometric variations in target areas within the images. Subsequently, a joint judgment mechanism based on the Structural Similarity Index (SSIM) and Mean Squared Error (MSE) is used to compare and analyze the images before and after rectification. SSIM mainly evaluates the perceptual similarity between two images from three aspects: brightness, contrast, and structure, reflecting the consistency of image structural information; MSE calculates the average difference between the two images from the perspective of pixel intensity. By weighted fusion of SSIM and MSE, the system can simultaneously consider visual structural changes and pixel errors, achieving a comprehensive quantification of image differences. Finally, the result judgment submodule outputs the rectification qualification judgment result based on the comprehensive index of image comparison. When the comparison result reaches a preset similarity threshold, the system automatically determines that the rectification is qualified; if it is below the threshold, a re-inspection or upgrade rectification process is triggered. The judgment result will be automatically transmitted to the knowledge optimization module to update the security rule knowledge base, thereby realizing the closed-loop accumulation and adaptive optimization of experience knowledge.

[0044] It can be seen that by verifying the multi-source image acquisition, deep feature comparison, and automatic judgment mechanism of the feedback module, intelligent, automated, and closed-loop management of hazard rectification at construction sites has been achieved. This module not only improves the accuracy and comprehensiveness of rectification effect evaluation, but also provides core technical support for knowledge optimization and continuous iteration in the PDCA-SDCA dual-cycle mechanism, significantly enhancing the intelligence level and reliability of railway engineering safety management.

[0045] Specifically, the knowledge optimization module is configured to update the rule entries in the security rule knowledge base based on the verification results and trigger a new round of PDCA process.

[0046] Specifically, the knowledge optimization module includes: a rule update submodule, configured to add, delete, and modify rule entries in the security rule knowledge base based on the results of the verification feedback module; a rule evaluation submodule, configured to evaluate the validity of rule entries through cross-validation algorithms and remove redundant rules; and a rule publishing submodule, configured to publish the updated rule entries to the rule reasoning module, triggering a new round of PDCA process.

[0047] Specifically, the rule publishing submodule uses an online random forest algorithm, which triggers a model update every 100 new verification data points.

[0048] Understandably, a dynamic rule learning and adaptive model update mechanism based on verification feedback enables the continuous evolution of the safety rule knowledge base and the construction of the system's self-learning capability. Its core idea is to use the results of hazard rectification verification as knowledge feedback signals. Through algorithm-driven rule optimization and model retraining, a closed-loop knowledge evolution system of "perception-feedback-learning-optimization" is formed, thereby achieving self-updating and sustainable iteration of safety management knowledge. First, the knowledge optimization module receives result data from the verification feedback module, including rectification pass rate, image similarity index, and deviation distribution characteristics. By comparing the difference between the rule execution effect and the actual rectification effect, it identifies inefficient, redundant, or outdated rules in the current safety rule base. Based on this, the rule update submodule performs addition, deletion, and modification operations on knowledge base entries. Its technical principle lies in adaptively adjusting rule entries according to the hazard type, triggering conditions, and rectification results in the verification feedback. For example, if a rule is triggered multiple times but the rectification results are consistently unsatisfactory, the system automatically determines that the rule's weight has decreased or its logic is unreasonable, and optimizes or deletes it. Conversely, when the rectification of a certain type of hidden danger is consistently efficient, the corresponding rule parameters are solidified and the priority is increased, thereby achieving dynamic evolution and stable convergence of the rule system. To ensure the reliability and scientific nature of updated rules, the rule evaluation submodule introduces a cross-validation algorithm to statistically evaluate the effectiveness of candidate rule entries. This module evaluates the adaptability and generalization performance of rules under different data distributions by dividing the validation dataset into training and test sets for multiple rounds of alternating validation, and automatically removes redundant or duplicate rule entries. Its technical principle lies in using model validation stability indicators (such as accuracy, F1 score, confidence interval, etc.) to measure the structural rationality of the rule set, so as to prevent logical conflicts or overfitting caused by frequent updates to the rule base. Once the rule system is stable, the rule publishing submodule is responsible for publishing the optimized rule entries to the rule reasoning module and triggering a new round of the PDCA process, realizing a continuous cycle of "Plan - Do - Check - Act". To further improve the intelligence level of rule reasoning, the rule publishing submodule builds a dynamic learning model based on the Online Random Forest algorithm, automatically triggering a model update every 100 new validation data points. This involves using an incremental learning mechanism to locally retrain the decision tree cluster, enabling the model to absorb new knowledge without complete retraining, thereby achieving real-time optimization of the rule matching model and continuous self-learning of the knowledge base.

[0049] As can be seen, through the dynamic updating, evaluation, and publishing functions of the knowledge optimization module, this invention achieves the continuous self-learning capability of the safety rule knowledge base. The system can continuously improve rule entries based on actual rectification results, enhance the accuracy of hazard identification and risk assessment, and ensure the efficient operation of the PDCA and SDCA cycles, thereby providing core technical support for the intelligent, full-process closed-loop management of railway engineering construction sites.

[0050] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principle of this invention is further explained below in conjunction with a specific application scenario.

[0051] At a railway construction site, suppose there is a potential hazard of improperly stacked materials in a bridge or culvert construction area. First, a drone inspection device collects image data of the area via its data acquisition module and transmits the data to the hazard analysis unit. The feature extraction submodule within the hazard analysis unit extracts key attributes from the image data, such as the location and height of the material stacking, and information about the surrounding environment. Subsequently, the feature mapping submodule maps these key attributes to a predefined feature space, forming standardized feature vectors. The feature fusion submodule weights and fuses these feature vectors from multiple dimensions to generate a comprehensive feature representation. The weighting coefficients are dynamically adjusted based on historical data statistical patterns to ensure the accuracy of the feature representation. The hazard analysis unit then transmits the comprehensive feature representation to the rule-based reasoning module. Specifically, the dynamic adjustment of the weighting coefficients based on historical data statistical patterns includes: dynamically adjusting the feature weights based on the information entropy weighting method combined with historical hazard data, using the following formula: .

[0052] The rule matching submodule within the rule reasoning module is based on a graph-structured safety rule knowledge base. It uses a depth-first search algorithm to find rule entries that match the current hazard's feature vector. This submodule supports fuzzy matching, allowing for a certain degree of feature deviation, thus improving matching flexibility. The risk assessment submodule calculates the hazard's risk level based on the matching results and generates corresponding warning signals. The instruction generation submodule generates rectification instructions based on the risk level and warning signals, and then transmits these instructions to the work order dispatch module.

[0053] The task queue management submodule within the work order assignment module prioritizes rectification tasks based on the risk level and occurrence time of the hazards. The task assignment submodule selects the optimal responsible party based on their responsibilities and geographical location, and pushes the rectification task to their terminal. The task status tracking submodule records the execution progress of rectification tasks in real time and synchronizes the status information to the verification feedback module.

[0054] After the responsible party completes the rectification, they acquire the rectified image data using drone re-capture and upload it to the verification feedback module. The image acquisition submodule within the verification feedback module receives the rectified image data, while the image comparison submodule uses a convolutional neural network to extract features and perform comparative analysis on the image data before and after rectification. The image comparison submodule supports multi-angle image fusion to improve comparison accuracy. The result judgment submodule determines whether the rectification is qualified based on the comparison results and transmits the judgment result to the knowledge optimization module.

[0055] The rule update submodule in the knowledge optimization module adds, deletes, and modifies rule entries in the security rule knowledge base based on the results of the verification feedback module. The rule evaluation submodule evaluates the validity of rule entries using a cross-validation algorithm and removes redundant rules. The rule publishing submodule publishes the updated rule entries to the rule reasoning module, triggering a new round of the PDCA process.

[0056] In the above steps, the data acquisition module obtains construction site information through drone inspection equipment, IoT sensors, and video surveillance devices, and stores the data in a data cache area. The data cache area is divided into a fixed cache area and a temporary cache area. The fixed cache area stores frequently acquired data, such as mechanical equipment operating parameters and environmental temperature and humidity, while the temporary cache area stores frequently acquired data, such as material stacking locations and construction process execution status. When the fixed cache area reaches its preset capacity, the temporary cache area triggers a data migration command, migrating some data to the temporary cache area. The temporary cache area responds to the data migration command and marks the timestamps of the migrated data as reference points. When the reference point reaches a preset time threshold, the temporary cache area triggers a data cleanup command, removing data exceeding the time threshold to free up storage space. The hazard analysis unit responds to the data cleanup command and recalculates the feature weight coefficients based on the remaining data. When the change in the feature weight coefficients exceeds a preset threshold, the hazard analysis unit triggers a re-matching command from the rule inference module to ensure the accuracy of the inference results.

[0057] As can be seen from the above steps, this invention achieves closed-loop management of the entire process of hazard identification, rectification, verification, and knowledge optimization. The hazard analysis unit, through the collaborative work of the feature extraction, feature mapping, and feature fusion sub-modules, realizes multi-dimensional analysis and weighted fusion of hazard features, improving the accuracy of hazard classification. The rule reasoning module, through the cooperation of the rule matching, risk assessment, and instruction generation sub-modules, realizes quantitative assessment of hazard risks and automatic generation of rectification instructions, improving the efficiency of hazard handling. The work order assignment module, through the collaboration of the task queue management, task assignment, and task status tracking sub-modules, realizes priority ranking and dynamic adjustment of rectification tasks, ensuring that high-risk hazards are handled first. The verification feedback module, through the cooperation of the image acquisition, image comparison, and result judgment sub-modules, realizes automated verification and accurate judgment of rectification effects, avoiding subjective errors caused by manual intervention. The knowledge optimization module, through the collaborative work of the rule update, rule evaluation, and rule release sub-modules, realizes dynamic updating of the safety rule knowledge base and removal of redundant rules, ensuring the effectiveness and applicability of the rules.

[0058] Furthermore, the collaborative operation of the data cache and temporary cache enables efficient data storage management and rational utilization of storage space, avoiding system performance degradation due to data overflow. By introducing the 5M1E feature deconstruction algorithm and a security rule knowledge base stored in a graph structure, this invention improves the accuracy and comprehensiveness of hazard identification. Through drone re-photographing and image comparison technology, automated verification of rectification effects is achieved. Incremental learning algorithms and version control mechanisms ensure the continuous optimization and traceability of the security rule knowledge base.

[0059] The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0060] In the above embodiments, the data acquisition module enables real-time acquisition of multimodal information from the construction site, including multi-source data such as drone inspections, IoT sensors, and video surveillance, thereby ensuring a comprehensive and accurate data foundation for hazard identification. Combined with the hazard analysis unit, which uses the 5M1E feature deconstruction algorithm to perform structured analysis of hazards from six dimensions—human, machine, material, method, environment, and measurement—it can standardize the representation of safety hazards from different construction specialties, improving the accuracy and systematization of hazard identification. Secondly, the rule reasoning module performs real-time scenario-based logical reasoning through a graph structure knowledge base, automating hazard classification and rectification instruction generation, reducing reliance on manual experience and improving the intelligence level of safety management. The work order assignment module automatically allocates rectification tasks to the responsible personnel's terminals, ensuring clear responsibility, efficient execution, and rapid implementation of rectification measures. Furthermore, the verification and feedback module automatically compares and analyzes the rectification effects using image data from drone re-captures or uploaded by the responsible personnel's terminals, achieving closed-loop management of the hazard handling process. Unsatisfactory rectifications are promptly upgraded, ensuring traceability and controllability throughout the entire safety management process. The knowledge optimization module continuously updates the safety rule knowledge base based on verification results, triggering a new round of PDCA processes, thereby achieving continuous self-learning of knowledge and improving the system's intelligent capabilities. In summary, this invention, through a PDCA-SDCA dual-cycle mechanism, achieves closed-loop management of the entire process of hazard identification, rectification, verification, and knowledge base optimization. It possesses significant advantages in terms of multi-dimensionality, automation, and intelligence, improving the efficiency of on-site safety management, the accuracy of early warnings, and the system's self-optimization capabilities, demonstrating high practicality and promotional value.

[0061] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0063] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A work order closed-loop management system based on PDCA-SDCA dual-cycle, characterized in that, include: The data acquisition module is configured to acquire construction site information based on drone inspection equipment, IoT sensors, and video surveillance devices. The hazard analysis unit is connected to the data acquisition module. The hazard analysis unit is equipped with several analysis channels, and each analysis channel corresponds to a different construction specialty. Based on the 5M1E feature deconstruction algorithm, the hazard is analyzed in a structured manner from multiple dimensions to generate standardized feature vectors. The rule reasoning module communicates with the hazard analysis unit. The rule reasoning module is configured to perform real-time scenario-based logical reasoning based on a safety rule knowledge base stored in a graph structure, determine the hazard classification, and generate rectification instructions. The work order dispatch module communicates with the rule reasoning module. The work order dispatch module is configured to receive rectification instructions and assign rectification tasks to the responsible person's terminal. The verification feedback module is connected to the data acquisition module and the responsible person's terminal. The verification feedback module is configured to automatically compare and analyze the rectification effect based on the image data uploaded by the drone or the responsible person's terminal, according to the CNN network. The knowledge optimization module is configured to update rule entries in the security rule knowledge base based on incremental learning algorithms and verification results, and trigger a new round of PDCA process.

2. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 1, characterized in that, The data acquisition module has a data buffer area, which is divided into a fixed buffer area and a temporary buffer area. A fixed buffer is used to store data acquired at high frequencies, while a temporary buffer is used to store data acquired at low frequencies. When the fixed cache reaches the preset capacity, the temporary cache triggers a data migration command to migrate some data to the temporary cache, where the preset capacity is 1TB. When the timestamp of the migrated data reaches a preset time threshold, the temporary cache triggers a data cleanup command to remove the data that exceeds the time threshold, which is 7 days.

3. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 2, characterized in that, The hidden danger analysis unit responds to the data cleaning command and dynamically recalculates the feature weight coefficients according to the information entropy weight method; When the change in the feature weight coefficient exceeds the preset threshold, the hidden danger analysis unit triggers a rematching instruction from the rule reasoning module.

4. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 1, characterized in that, The hazard analysis unit includes: The feature extraction submodule is connected to the data acquisition module and is configured to extract key attributes from the raw data. The feature mapping submodule is connected to the feature extraction submodule. The feature mapping submodule is configured to map the extracted key attributes to a predefined feature space to form a standardized feature vector. The feature fusion submodule communicates with the feature mapping submodule. The feature fusion submodule is configured to perform weighted fusion of feature vectors from multiple dimensions to generate a comprehensive feature representation. The weight coefficients are dynamically adjusted based on historical data statistical patterns.

5. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 1, characterized in that, The rule-based reasoning module includes: The rule matching submodule is configured as a security rule knowledge base stored based on a graph structure. The rule matching submodule is also configured to search for rule entries that match the current hazard feature vector based on a depth-first search algorithm. The risk assessment submodule is configured to calculate the risk level of potential hazards based on the matching results and generate corresponding early warning signals; The instruction generation submodule is configured to generate rectification instructions based on risk level and early warning signals, and then transmit the instructions to the work order dispatch module.

6. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 1, characterized in that, The work order dispatch module includes: The task queue management submodule is configured to prioritize rectification tasks according to the risk level and occurrence time of the hazards; The task assignment submodule is configured to select the optimal responsible person based on their responsibilities and geographical location information, and then push the rectification task to the responsible person's terminal. The task status tracking submodule is configured to record the execution progress of rectification tasks in real time and synchronize the status information to the verification feedback module. The responsible person's terminal is a smart handheld device running the Android system, which supports work order reception, image uploading, and progress feedback.

7. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 1, characterized in that, The verification feedback module includes: The image acquisition submodule is configured to acquire the rectified image data through drone re-shooting or uploading via the responsible person's terminal; The image alignment submodule is configured to extract image features based on a pre-trained ResNet-50 model. The image alignment submodule is also configured to comprehensively determine the rectification effect using the structural similarity index (SSIM) and mean squared error (MSE). The result determination submodule is configured to determine whether the rectification is qualified based on the comparison results and pass the determination results to the knowledge optimization module.

8. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 1, characterized in that, The knowledge optimization module includes: The rule update submodule is configured to add, delete, and modify rule entries in the security rule knowledge base based on the results of the verification feedback module. The rule evaluation submodule is configured to evaluate the validity of rule entries using a cross-validation algorithm and remove redundant rules. The rule publishing submodule is configured to publish updated rule entries to the rule reasoning module, triggering a new round of PDCA process.

9. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 3, characterized in that, The rule reasoning module responds to the rematching instruction and recalculates the risk level of the hazard based on the updated feature weight coefficients; When the risk level changes, the rule reasoning module triggers the task priority adjustment instruction from the work order assignment module.

10. The work order closed-loop management system based on PDCA-SDCA dual-cycle as described in claim 2, characterized in that, The temporary cache responds to data migration instructions and marks the timestamps of the migrated data as reference points. When the reference point reaches a preset time threshold, the temporary cache triggers a data cleanup instruction to remove data that exceeds the time threshold in order to free up storage space.

Images