Intelligent maintenance decision-making system and method for subway train in full life cycle

Abstract

The invention relates to the crossing field of rail transit intelligent operation and maintenance and digital twinning technologies, and discloses a subway train full-life-cycle intelligent maintenance decision-making system and a subway train full-life-cycle intelligent maintenance decision-making method. The system comprises a digital twin modeling module used for constructing a train full-dimension virtual model and realizing real-time data synchronization with a physical entity; the full-life-cycle data acquisition and fusion module is used for integrating multi-source heterogeneous data of all stages of design, manufacturing, operation, maintenance and decommissioning to form a unified data asset library; the intelligent analysis and decision-making module is used for performing state monitoring, fault prediction and residual life evaluation based on the virtual model and the data asset library, and generating a maintenance decision-making scheme through multi-objective optimization; and the maintenance execution and feedback module is used for issuing a decision and executing data feedback to realize closed-loop optimization of the system. The method is realized based on the system. The method can support the subway train maintenance mode-to-state guide conversion.

Description

Technical Field

[0001] This invention relates to the intersection of intelligent operation and maintenance of rail transit and digital twin technology, specifically to an intelligent maintenance decision-making system and method for the entire life cycle of subway trains based on digital twins. Background Technology

[0002] Subway trains are the core mode of urban public transportation, and their operational safety and reliability are of paramount importance. Currently, the maintenance of subway trains generally adopts a planned preventive maintenance model based on fixed cycles. This model relies heavily on manual experience to formulate maintenance plans, which has significant limitations: on the one hand, it may lead to "over-maintenance," that is, unnecessary disassembly, inspection, or replacement of components when they are still in good condition, resulting in high operation and maintenance costs; on the other hand, it may also lead to "under-maintenance," that is, the failure to identify potential faults in a timely manner, thereby bringing operational safety risks.

[0003] With the development of information technology, subway trains generate massive amounts of heterogeneous data from multiple sources during design, manufacturing, operation, and maintenance. However, this data is usually scattered across different subsystems, forming "data silos" and lacking effective integration and correlation analysis methods, making it difficult to trace the entire lifecycle status and deeply mine its value. In addition, existing maintenance systems lack real-time and accurate mapping of the physical train status, and the virtual model is severely disconnected from the physical entity, resulting in delayed maintenance decisions and an inability to achieve proactive fault prediction and health warnings.

[0004] Digital twin technology, as a key technology for realizing two-way mapping and interaction between the physical world and cyberspace, has been initially explored in the industrial field. However, when applied to intelligent maintenance of subway trains throughout their entire lifecycle, existing solutions often focus on single aspects such as status monitoring or model display, and have not yet formed a complete technical solution and system implementation integrating "real-time mapping, data fusion, intelligent decision-making, and closed-loop optimization." This makes it difficult to build a collaborative closed loop from data acquisition to decision execution and feedback, and fails to meet the comprehensive needs of the subway industry for high safety, high economy, and high timeliness in maintenance operations.

[0005] Therefore, there is an urgent need in this field for a systematic solution that can integrate data throughout the entire lifecycle and achieve accurate state perception and intelligent decision optimization. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a digital twin-based intelligent maintenance decision-making system and method for the entire life cycle of subway trains, so as to realize the transformation from fixed cycle maintenance to condition-oriented maintenance and improve the accuracy, foresight and economy of maintenance decisions.

[0007] To solve the above-mentioned technical problems, the present invention is implemented as follows: This invention provides an intelligent maintenance decision-making system for the entire lifecycle of subway trains, comprising: The digital twin modeling module is used to construct a full-dimensional virtual model covering the geometric, physical, and performance attributes of key components of subway trains, and to establish a real-time data synchronization link between the physical entity and the virtual model to support the virtual model to be iteratively updated based on full lifecycle data. The full lifecycle data acquisition and fusion module is used to collect multi-source heterogeneous data generated by subway trains during the design, manufacturing, operation, maintenance and decommissioning stages, and to preprocess and fuse the multi-source heterogeneous data to form a unified format full lifecycle data asset library. The intelligent analysis and decision-making module, based on the virtual model status output by the digital twin modeling module and the full life cycle data asset library, performs status monitoring, fault prediction and remaining life assessment, and generates maintenance decision schemes with operational safety, maintenance cost and operational efficiency as optimization objectives through a multi-objective optimization algorithm based on the assessment results. The maintenance execution and feedback module is used to distribute the maintenance decision plan to the maintenance terminal and collect the actual data generated during the maintenance execution process and feed it back to the system. It is used to perform closed-loop optimization of the model parameters of the digital twin modeling module and the algorithm logic of the intelligent analysis and decision module. The visualization and interaction module is used to provide a linked visual display of the virtual model and the physical entity's state, interactive operation, and maintenance decision report generation functions; and Multiple onboard sensors deployed on key components of the train are used to collect status data during train operation.

[0008] Furthermore, the key components include the train body, bogies, traction system, and braking system.

[0009] Furthermore, the multi-source heterogeneous data includes real-time operating data, maintenance record data, historical fault data, design and manufacturing data, and decommissioning assessment data collected by the vehicle-mounted sensors.

[0010] Furthermore, in the intelligent analysis and decision-making module: The fault prediction is implemented using a Long Short-Term Memory (LSTM) neural network; The remaining lifetime assessment is implemented using a combination of rainflow counting and random forest algorithms. The maintenance decision scheme is generated using a multi-objective genetic algorithm.

[0011] Furthermore, the vehicle-mounted sensors include vibration sensors, temperature sensors, pressure sensors, and speed sensors.

[0012] Furthermore, the visualization and interaction module supports fault simulation and troubleshooting plan rehearsal functions.

[0013] Furthermore, the full lifecycle data acquisition and fusion module uses a weighted fusion algorithm or a deep learning-based fusion algorithm to integrate the multi-source heterogeneous data.

[0014] Furthermore, the system also includes edge computing nodes and a cloud platform, wherein: The edge computing nodes are deployed on the train and are used to perform real-time preprocessing of the data collected by the on-board sensors. The cloud platform is used to host the digital twin modeling module, the full lifecycle data asset library, and the intelligent analysis and decision-making module.

[0015] This invention also provides a digital twin-based intelligent maintenance decision-making method for the entire lifecycle of subway trains, applied to the aforementioned system, the method comprising: By deploying onboard sensors on key train components, the status data of the subway train during operation is collected; Construct a full-dimensional virtual model of key components of subway trains and establish a real-time data synchronization link between physical entities and virtual models; Collect and integrate multi-source heterogeneous data generated during the design, manufacturing, operation, maintenance and decommissioning stages of subway trains to form a unified format full life cycle data asset library; Based on the state of the virtual model and the full lifecycle data asset library, perform status monitoring, fault prediction and remaining life assessment. Based on the results of fault prediction and remaining life assessment, a maintenance decision scheme with operational safety, maintenance cost and operational efficiency as optimization objectives is generated through a multi-objective optimization algorithm. The maintenance decision plan is sent to the maintenance terminal for execution, and the actual data generated during the maintenance execution process is collected. Using the maintenance execution data, the parameters of the virtual model and the logic of the multi-objective optimization algorithm are optimized in a closed-loop iterative manner.

[0016] Furthermore, the fault prediction employs a Long Short-Term Memory (LSTM) neural network with an attention mechanism; the remaining lifetime assessment is achieved using a rainflow counting method combined with a random forest algorithm; and the multi-objective optimization algorithm is a multi-objective genetic algorithm.

[0017] Compared with the prior art, the advantages of this invention are as follows: (1) Realized the integration and fusion of data throughout the entire life cycle: Through multi-source data collection and fusion technology, information silos were broken down, providing a comprehensive data foundation for accurate decision-making.

[0018] (2) It has promoted the transformation of maintenance mode: by using the real-time mapping and driving of digital twin technology, it has realized the transformation from the traditional "fixed cycle maintenance" to the advanced "condition-oriented maintenance", effectively avoiding over-maintenance and under-maintenance.

[0019] (3) Improved the intelligence and foresight of decision-making: By introducing advanced machine learning algorithms, accurate prediction of failure risk and remaining life of components was achieved, which significantly improved the scientific nature of decision-making and reduced operational safety risks.

[0020] (4) Optimized maintenance costs and operational efficiency: By optimizing the timing, content and resource allocation of maintenance, maintenance costs were reduced while ensuring safety, and train operation efficiency was improved by reducing unnecessary downtime.

[0021] (5) It gives the system the ability to self-evolve: Based on the closed-loop optimization mechanism, the system can continuously improve itself by using feedback data, so that the model accuracy and decision-making ability can continuously evolve with the use process and adapt to the state changes of the train throughout its entire life cycle.

[0022] Furthermore, the system's closed-loop optimization mechanism, implemented through the maintenance execution and feedback module, endows the system with continuous self-evolution capabilities. Specifically, the system utilizes actual data fed back after each maintenance execution (such as actual maintenance duration, replaced component models, and test data) to calibrate the performance degradation parameters of the digital twin model and iteratively optimize the parameters and logic of intelligent analysis algorithms (such as LSTM, random forest, and multi-objective genetic algorithms). This mechanism enables the system to continuously adapt to changes in the train's actual condition during use, thereby continuously improving the accuracy and adaptability of fault prediction, life assessment, and maintenance decisions. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 The overall structure diagram of the intelligent maintenance decision-making system for the entire life cycle of subway trains based on digital twins provided by this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The terms "first," "second," etc., used in the specification and claims of this invention are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0026] This invention provides a digital twin-based intelligent maintenance decision-making system for the entire lifecycle of subway trains, including a digital twin modeling module, a lifecycle data acquisition and fusion module, an intelligent analysis and decision-making module, a visualization and interaction module, and multiple onboard sensors.

[0027] The digital twin modeling module is used to construct a full-dimensional virtual model covering the geometric, physical, and performance attributes of key components of subway trains, and to establish a real-time data synchronization link between the physical entity and the virtual model to support the iterative updates of the virtual model based on full lifecycle data.

[0028] The full lifecycle data acquisition and fusion module is used to collect multi-source heterogeneous data generated by subway trains during the design, manufacturing, operation, maintenance and decommissioning stages, and to preprocess and fuse the multi-source heterogeneous data to form a unified format full lifecycle data asset library.

[0029] The full lifecycle data acquisition and fusion module uses a weighted fusion algorithm or a deep learning-based fusion algorithm to integrate the multi-source heterogeneous data.

[0030] The multi-source heterogeneous data includes real-time operating data, maintenance record data, historical fault data, design and manufacturing data, and decommissioning assessment data collected by the vehicle-mounted sensors.

[0031] The intelligent analysis and decision-making module, based on the virtual model status output by the digital twin modeling module and the full lifecycle data asset library, performs status monitoring, fault prediction and remaining life assessment, and generates maintenance decision schemes with operational safety, maintenance cost and operational efficiency as optimization objectives through a multi-objective optimization algorithm based on the assessment results.

[0032] In the intelligent analysis and decision-making module: The fault prediction is implemented using a Long Short-Term Memory (LSTM) neural network; The remaining lifetime assessment is implemented using a combination of rainflow counting and random forest algorithms. The maintenance decision scheme is generated using a multi-objective genetic algorithm.

[0033] The maintenance execution and feedback module is used to distribute the maintenance decision plan to the maintenance terminal and collect the actual data generated during the maintenance execution process and feed it back to the system. It is used to perform closed-loop optimization of the model parameters of the digital twin modeling module and the algorithm logic of the intelligent analysis and decision module. The visualization and interaction module is used to provide a linked visualization display of the virtual model and the physical entity's state, interactive operation, and maintenance decision report generation functions.

[0034] The visualization and interaction module supports fault simulation and troubleshooting plan rehearsal functions.

[0035] Multiple onboard sensors are deployed on key components of the train to collect status data during train operation. These key components include the train body, bogies, traction system, and braking system.

[0036] Furthermore, the vehicle-mounted sensors include vibration sensors, temperature sensors, pressure sensors, and speed sensors.

[0037] The system also includes edge computing nodes and a cloud platform, wherein: The edge computing nodes are deployed on the train and are used to perform real-time preprocessing of the data collected by the on-board sensors. The cloud platform is used to host the digital twin modeling module, the full lifecycle data asset library, and the intelligent analysis and decision-making module.

[0038] This invention also provides a digital twin-based intelligent maintenance decision-making method for the entire lifecycle of subway trains, applied to the aforementioned system, the method comprising: By deploying onboard sensors on key train components, the status data of the subway train during operation is collected; Construct a full-dimensional virtual model of key components of subway trains and establish a real-time data synchronization link between physical entities and virtual models; Collect and integrate multi-source heterogeneous data generated during the design, manufacturing, operation, maintenance and decommissioning stages of subway trains to form a unified format full life cycle data asset library; Based on the state of the virtual model and the full lifecycle data asset library, perform status monitoring, fault prediction and remaining life assessment. Based on the results of fault prediction and remaining life assessment, a maintenance decision scheme with operational safety, maintenance cost and operational efficiency as optimization objectives is generated through a multi-objective optimization algorithm. The maintenance decision plan is sent to the maintenance terminal for execution, and the actual data generated during the maintenance execution process is collected. Using the maintenance execution data, the parameters of the virtual model and the logic of the multi-objective optimization algorithm are optimized in a closed-loop iterative manner.

[0039] Furthermore, the fault prediction employs a Long Short-Term Memory (LSTM) neural network with an attention mechanism; the remaining lifetime assessment is achieved using a rainflow counting method combined with a random forest algorithm; and the multi-objective optimization algorithm is a multi-objective genetic algorithm.

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiment 1. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0041] Example 1 The architecture of the system of this invention can be understood from two complementary perspectives: functional logic architecture and physical deployment architecture.

[0042] Functional logic architecture (e.g.) Figure 1 As shown in the diagram, this perspective focuses on the data flow and functional module division within the system, including, from bottom to top, the data acquisition layer, digital twin layer, intelligent decision-making layer, and application layer. This architecture diagram clearly defines the logical relationships and data interaction paths between the core modules (such as data fusion, virtual modeling, intelligent analysis, and visualization).

[0043] Physical deployment architecture (hereinafter referred to as "cloud-edge-device" collaborative architecture): This perspective focuses on the distribution and collaboration of system hardware and software components in the actual physical environment (such as trains, data centers, and operation and maintenance terminals), including the terminal perception layer (onboard sensors and acquisition terminals), the edge computing layer (onboard edge nodes), the cloud platform layer (central server cluster), and the interactive display layer (various terminal devices).

[0044] The two architectures are essentially unified: each module in the functional logic architecture is specifically deployed in the corresponding layer of the "cloud-edge-device" physical architecture based on its computing needs, real-time requirements, and data security. For example, the "data acquisition layer" function is mainly implemented by the terminal perception layer, while the core computing of the "intelligent decision-making layer" and "digital twin layer" is deployed on the cloud platform layer. The following section will elaborate on the specific implementation of the system based on the "cloud-edge-device" physical deployment architecture.

[0045] I. System Architecture and Hardware Deployment This system adopts a collaborative architecture of "cloud-edge-device", such as Figure 1 As shown, this is to ensure the real-time performance of data processing and the high reliability of the overall system. The specific hardware deployment is as follows: 1. Terminal perception layer (vehicle deployment): Sensor Deployment: Various industrial-grade sensors are deployed on key components of each subway train. Specifically: Vibration sensors (example model: PCB 352C33, with vibration resistance ≥10g) are installed on the bogie and traction motor to collect mechanical vibration signals.

[0046] Temperature sensors (example model: PT1000, range -40℃~125℃) are installed in the electrical cabinet and bearing housing to monitor temperature changes at key points.

[0047] Install a pressure sensor (example model: Honeywell ASPC series) in the brake line to monitor the pressure status of the brake system.

[0048] Install a speed sensor (example model: Omron E2B-M18KN16-WZ-B1) on the vehicle side to collect speed information of rotating parts such as wheels.

[0049] Data Acquisition: The sensor data mentioned above is aggregated and initially formatted by a data acquisition terminal (example model: NI-CDAQ-9178) deployed in the electrical cabinet of the train car. All onboard hardware meets rail transit industry standards (such as EN50159) and has vibration resistance, electromagnetic interference resistance, and wide temperature adaptability (-40℃~70℃).

[0050] 2. Edge computing layer: One edge computing node (example model: NVIDIA Jetson Xavier NX) is deployed in the control cabinets of each of the first and last carriages of the train. This node is responsible for local preprocessing of sensor data, including data cleaning, noise reduction, and feature extraction, to reduce the communication and computing burden on the cloud. At the same time, it provides real-time edge alarms for key anomalies (such as instantaneous vibration values ​​exceeding preset thresholds).

[0051] 3. Cloud Platform Layer: The system is centrally deployed in the operations and maintenance center's server room, with all hardware configured redundantly to ensure uninterrupted 24 / 7 operation. This layer includes: An application server cluster (example model: Dell PowerEdge R750, 4-unit cluster deployment) is used to host the system's core business logic.

[0052] A database server (example model: IBM System x3850 X6, 2 primary and backup redundant units) is used to store and manage the full lifecycle data asset repository.

[0053] The digital twin modeling server (example model: Huawei TaiShan 200 server, 2 units) runs a high-fidelity virtual model and its real-time data synchronization and mapping engine.

[0054] AI computing server: Equipped with AI acceleration cards (example model: NVIDIA A100, 4 cards), specifically designed to accelerate the training and inference process of algorithms such as LSTM and random forest, and control the real-time computing latency of key algorithms to within 500ms.

[0055] Network communication is ensured by industrial switches (example model: Cisco Catalyst 9300 series).

[0056] 4. Interactive Presentation Layer: Equip train drivers and maintenance personnel with industrial-grade tablets (example model: Huawei MatePad Pro).

[0057] Set up a monitoring dashboard (example: Samsung 55-inch video wall, 3×2 layout) in the operations and maintenance center for global status visualization.

[0058] Desktop computers (example model: Lenovo ThinkCentre M900) are deployed at workstations in the maintenance workshop to receive maintenance decisions, provide feedback on execution data, and enable visual interaction.

[0059] II. System Workflow A complete workflow example of this system is as follows: Data collection and uploading: During train operation, onboard sensors continuously collect status data such as vibration, temperature, pressure, and rotational speed. After the data acquisition terminal aggregates this data, edge computing nodes perform preprocessing operations such as data cleaning, noise reduction, and extraction of time-domain features (e.g., peak value, mean, and variance). The preprocessed data is then uploaded to the cloud platform in real time via 5G / wireless network. Simultaneously, maintenance records and fault information entered by maintenance personnel through terminals, as well as historical data imported from the design and manufacturing stages, are also synchronously entered and stored in the cloud database.

[0060] Data fusion and model-driven approaches: The cloud-based full lifecycle data acquisition and fusion module employs a deep learning-based fusion algorithm to associate and integrate uploaded real-time data with historical multi-source heterogeneous data, storing it in a unified data format full lifecycle data asset library. Subsequently, the digital twin modeling module calls upon the data in this library to drive the dynamic updating of the virtual model, achieving accurate and real-time mapping of the physical train's operating status and health condition.

[0061] Intelligent analysis and decision generation: The intelligent analysis and decision-making module is run periodically or triggered by events.

[0062] Fault Prediction: The fault prediction unit inputs the processed vibration time-domain features and temperature data into a pre-trained improved LSTM neural network model (introducing an attention mechanism and setting an appropriate number of neurons in the hidden layers). The model outputs a specific fault type (such as bearing wear or gear breakage) and its probability of occurrence within the next 72 hours. If the probability exceeds a preset threshold (e.g., 0.85), a system alarm is triggered.

[0063] Remaining life assessment: The remaining life assessment unit targets alarm components and, in combination with their operating time, load spectrum calculated by rainflow counting (used to calculate cumulative damage), and maintenance records, uses a random forest model (50 decision trees) to predict their remaining life range (e.g., 1200-1500 hours) with a confidence level of over 95%.

[0064] Maintenance Decision Optimization: The maintenance decision optimization unit uses multiple objectives—"lowest safety risk, lowest maintenance cost, and shortest downtime"—and employs a multi-objective genetic algorithm (population size 50, 50 iterations) to comprehensively consider fault prediction results, remaining lifespan, existing maintenance resources (personnel, spare parts), and operational plans (maintenance windows) to generate the optimal maintenance plan. For example: "It is recommended to replace the left traction motor bearing of the bogie of car 03 within 3 days in the D12 maintenance tunnel. One senior technician is required, and the work time is approximately 4 hours."

[0065] Decision execution and closed-loop feedback: The generated maintenance plan is pushed to the tablets of relevant maintenance personnel and PCs at their workstations in the workshop via the system. After executing the plan, maintenance personnel provide feedback through their terminals on the actual maintenance time, the models of replaced parts, and post-maintenance test data. This maintenance execution data is collected by the system and sent back to the full lifecycle data asset library. The closed-loop optimization unit of the maintenance execution and feedback module uses this feedback data to fine-tune the parameters of the LSTM model and the random forest model, and correct the performance degradation parameters in the digital twin model, thereby completing an iterative optimization and continuously improving the accuracy of the next decision.

[0066] III. Implementation Examples of Core Algorithms Fault prediction algorithm (improved LSTM): First, the input multi-source data such as vibration and temperature are processed by Min-Max normalization, and the Pearson correlation coefficient method is used for feature selection to remove redundant features.

[0067] Subsequently, the dataset was divided into training and testing sets in a 7:3 ratio.

[0068] During model training, the learning rate was set to 0.001, and the iterations were 1000. An attention mechanism was introduced after the LSTM layer, enabling the model to focus on features at key time steps and improve prediction accuracy.

[0069] A model can only be used for actual predictions if its prediction accuracy on the test set reaches a preset accuracy threshold.

[0070] Remaining lifetime assessment algorithm (rainflow counting method + random forest): First, the rainflow counting method is used to perform cyclic counting of the component load time series, and the cumulative damage degree is calculated by combining the material's SN curve.

[0071] Then, damage accumulation rate, average load, operating environment temperature, and recent maintenance frequency are used as features to construct a feature dataset.

[0072] The model was trained using the random forest algorithm (with 50 decision trees).

[0073] The error in the remaining lifetime prediction range output by the model should be controlled within 8%.

[0074] Maintenance decision optimization algorithm (multi-objective genetic algorithm): The maintenance cost, safety risk value (calculated from the probability and severity of failure), and expected downtime are constructed as a multi-objective optimization function.

[0075] Constraints include the number of available technicians, spare parts inventory, and train operation windows.

[0076] Fifty random maintenance schemes were initialized as the initial population, and iterative evolution was carried out through roulette wheel selection, single-point crossover (crossover probability 0.8), and basic bit mutation (mutation probability 0.05).

[0077] Finally, the optimal maintenance scheme that satisfies all constraints is selected by using the Pareto optimal solution front.

[0078] The system described in this invention, through the above-mentioned hardware and software collaborative implementation scheme, realizes intelligent, accurate and efficient maintenance decision support for subway trains throughout their entire life cycle, from design and operation to decommissioning.

[0079] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0080] Furthermore, it should be noted that the scope of the methods and systems in the embodiments of the present invention is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. In addition, features described with reference to certain examples may be combined in other examples.

[0081] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A smart maintenance decision-making system for the entire life cycle of subway trains, characterized in that, include: The digital twin modeling module is used to construct a full-dimensional virtual model covering the geometric, physical, and performance attributes of key components of subway trains, and to establish a real-time data synchronization link between the physical entity and the virtual model to support the virtual model to be iteratively updated based on full lifecycle data. The full lifecycle data acquisition and fusion module is used to collect multi-source heterogeneous data generated by subway trains during the design, manufacturing, operation, maintenance and decommissioning stages, and to preprocess and fuse the multi-source heterogeneous data to form a unified format full lifecycle data asset library. The intelligent analysis and decision-making module, based on the virtual model status output by the digital twin modeling module and the full life cycle data asset library, performs status monitoring, fault prediction and remaining life assessment, and generates maintenance decision schemes with operational safety, maintenance cost and operational efficiency as optimization objectives through a multi-objective optimization algorithm based on the assessment results. The maintenance execution and feedback module is used to distribute the maintenance decision plan to the maintenance terminal and collect the actual data generated during the maintenance execution process and feed it back to the intelligent maintenance decision system for the entire life cycle of the metro train. It is used to perform closed-loop optimization of the model parameters of the digital twin modeling module and the algorithm logic of the intelligent analysis and decision module. The visualization and interaction module is used to provide a linked visualization display of the virtual model and the physical entity's state, interactive operation, and maintenance decision report generation functions; as well as Multiple onboard sensors deployed on key components of the train are used to collect status data during train operation.

2. The system according to claim 1, characterized in that, The key components include the train body, bogies, traction system, and braking system.

3. The system according to claim 1, characterized in that, The multi-source heterogeneous data includes real-time operating data, maintenance record data, historical fault data, design and manufacturing data, and decommissioning assessment data collected by the vehicle-mounted sensors.

4. The system according to claim 1, characterized in that, In the intelligent analysis and decision-making module: The fault prediction is implemented using a long short-term memory neural network. The remaining lifetime assessment is implemented using a combination of rainflow counting and random forest algorithms. The maintenance decision scheme is generated using a multi-objective genetic algorithm.

5. The system according to claim 1, characterized in that, The vehicle-mounted sensors include vibration sensors, temperature sensors, pressure sensors, and speed sensors.

6. The system according to claim 1, characterized in that, The visualization and interaction module supports fault simulation and troubleshooting plan rehearsal functions.

7. The system according to claim 1, characterized in that, The full lifecycle data acquisition and fusion module uses a weighted fusion algorithm or a deep learning-based fusion algorithm to integrate the multi-source heterogeneous data.

8. The system according to claim 1, characterized in that, The system also includes edge computing nodes and a cloud platform, wherein: The edge computing nodes are deployed on the train and are used to perform real-time preprocessing of the data collected by the on-board sensors. The cloud platform is used to host the digital twin modeling module, the full lifecycle data asset library, and the intelligent analysis and decision-making module.

9. A method for intelligent maintenance decision-making throughout the entire life cycle of subway trains, characterized in that, Applied to the system as described in any one of claims 1-8, the method comprises: By deploying onboard sensors on key train components, the status data of the subway train during operation is collected; Construct a full-dimensional virtual model of key components of subway trains and establish a real-time data synchronization link between physical entities and virtual models; Collect and integrate multi-source heterogeneous data generated during the design, manufacturing, operation, maintenance and decommissioning stages of subway trains to form a unified format full life cycle data asset library; Based on the state of the virtual model and the full lifecycle data asset library, perform status monitoring, fault prediction and remaining life assessment. Based on the results of fault prediction and remaining life assessment, a maintenance decision scheme with operational safety, maintenance cost and operational efficiency as optimization objectives is generated through a multi-objective optimization algorithm. The maintenance decision plan is sent to the maintenance terminal for execution, and the actual data generated during the maintenance execution process is collected. Using the maintenance execution data, the parameters of the virtual model and the logic of the multi-objective optimization algorithm are optimized in a closed-loop iterative manner.

10. The method according to claim 9, characterized in that, The fault prediction adopts a long short-term memory neural network with an attention mechanism; the remaining lifetime assessment is implemented by rainflow counting combined with random forest algorithm; the multi-objective optimization algorithm is a multi-objective genetic algorithm.

Images