Dynamic health degree evaluation and predictive maintenance method for power equipment

Abstract

The invention discloses a power equipment dynamic health degree assessment and predictive maintenance method, and belongs to the technical field of railway power system operation and maintenance. The method comprises the following steps: constructing a parameterized digital twinborn body of power equipment, and collecting real-time operation data, resume data and environment data; based on the parameterized digital twins and the collected data, equipment health degree components are calculated through a multi-model cooperation method, and a comprehensive health index is generated through fusion; performing equipment life prediction and maintenance decision generation according to the comprehensive health index, and outputting an optimal maintenance strategy; and performing visual virtual rehearsal and augmented reality auxiliary execution on the optimal maintenance strategy to form a closed-loop maintenance system. According to the method, the problems of data and model separation, model static stiffness and health assessment deficiency in the prior art are solved, dynamic perception, accurate assessment and predictive maintenance of the equipment state are realized, and the operation and maintenance efficiency and the system reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of railway power system operation and maintenance technology, and in particular to methods for dynamic health assessment and predictive maintenance of power equipment. Background Technology

[0002] As a critical national infrastructure, the stable operation of the railway system highly depends on a continuous and reliable power supply. To ensure the safe and efficient operation and maintenance of railway power equipment, the industry has developed a management paradigm that combines multiple technical approaches.

[0003] First, Supervisory Control and Data Acquisition (SCADA) systems have been widely deployed. These systems collect real-time parameters such as current, voltage, temperature, and switch status of power equipment during operation, displaying and triggering alarms on a centralized monitoring interface, providing maintenance personnel with crucial real-time status awareness capabilities.

[0004] Secondly, to manage equipment's entire lifecycle data, equipment history databases have gradually replaced traditional paper records. This database systematically records basic equipment information, historical operation and maintenance records, faults, and repair information, providing valuable historical data support for maintenance decisions.

[0005] Recently, with the development of information visualization technology, 3D model display technology has begun to be introduced into the operation and maintenance field. Current practices typically involve synchronously displaying the 3D model of power equipment with the SCADA monitoring interface and equipment history information through side-by-side windows or split-screen displays, aiming to provide operation and maintenance personnel with a more intuitive reference.

[0006] However, existing technical solutions still have the following significant drawbacks:

[0007] Data and models are deeply disconnected and lack intrinsic connection: Most current systems use simple side-by-side interfaces to display data, but there is a lack of deep, structured logical connection between SCADA data, historical data, and 3D models. Maintenance personnel find it difficult to quickly and accurately map monitoring data streams (such as multiple ammeter readings) to the corresponding specific components in the 3D model. This is especially true for personnel unfamiliar with the layout of field equipment, making it even more difficult to establish this data-to-entity mapping relationship and easily leading to misjudgments.

[0008] 3D models are static and rigid, unable to dynamically reflect the real state: Most existing 3D models are static displays, only showing the basic geometric shape of the equipment. They cannot dynamically update their status based on real-time SCADA data. For example, the open / closed position of switches, the dynamic changes in ammeter readings, and the actual position of circuit breaker indicators cannot be visually reflected in the model in real time, limiting the value of 3D models to a rough appearance reference.

[0009] Lack of health assessment capabilities and reliance on experience for maintenance decisions: Existing technologies focus on data presentation and recording, lacking in-depth analysis and quantitative assessment of equipment health status. Maintenance decisions heavily depend on human experience in interpreting SCADA alarm thresholds and records, failing to provide early warnings of equipment performance degradation and scientific predictions of remaining lifespan. This results in maintenance strategies often being reactive or performed at fixed intervals, which is neither economical nor safe.

[0010] Inefficient model building and difficulty in balancing accuracy and cost: There are two common problems in 3D model building: First, directly using open-source similar models from the Internet, which leads to deviations between the model and the actual equipment on site in terms of structure and details; Second, relying entirely on manual modeling based on photos, which, although highly accurate, is time-consuming, labor-intensive, and costly, making it difficult to apply on a large scale.

[0011] Therefore, there is an urgent need in this field for an intelligent solution that can deeply integrate multidimensional data and high-precision models, and enable dynamic assessment of equipment health status and predictive maintenance decisions. Summary of the Invention

[0012] The purpose of this invention is to provide a method for dynamic health assessment and predictive maintenance of power equipment, in order to solve the aforementioned problems existing in the prior art.

[0013] The technical solution, a method for dynamic health assessment and predictive maintenance of power equipment, includes the following steps:

[0014] Construct a parameterized digital twin of power equipment and collect real-time operation data, historical data and environmental data;

[0015] Based on the parameterized digital twin and the collected data, the device health component is calculated using a multi-model collaborative method and then integrated to generate a comprehensive health index.

[0016] Based on the comprehensive health index, equipment lifespan is predicted and maintenance decisions are generated, and the optimal maintenance strategy is output.

[0017] The optimal maintenance strategy is visualized and virtually simulated, and then executed with augmented reality assistance to form a closed-loop maintenance system.

[0018] According to a further improvement of the present invention, constructing a parameterized digital twin of a power equipment includes:

[0019] Collect equipment design drawings, material property data, and rated parameter data, perform parametric modeling, and obtain a parametric 3D model;

[0020] It integrates a multiphysics simulation module, configures thermodynamic simulation parameters, electrical stress simulation parameters, and mechanical stress simulation parameters, and constructs a physical simulation model;

[0021] Collect equipment technical manuals, operation and maintenance procedures, historical fault records, and expert experience rules, perform natural language processing and rule extraction, construct an operation and maintenance knowledge graph, and form a knowledge model.

[0022] According to a further improvement of the present invention, real-time operational data, historical data, and environmental data are collected, including:

[0023] The system collects device current data, voltage data, temperature data, and switch status data through a data interface.

[0024] Query and collect maintenance record data, historical load data, and past fault data of the target equipment from the equipment history database;

[0025] It collects ambient temperature and humidity data reported by sensors deployed on-site, as well as inspection images and recorded data uploaded by inspection personnel via mobile terminals.

[0026] According to a further improvement of the present invention, the device health component is calculated using a multi-model collaborative method, including:

[0027] The driving physical simulation model uses real-time operating data of the SCADA system and environmental data as boundary conditions to perform multi-physics simulation calculations, obtain simulated temperature field distribution data, simulated electric stress distribution data and simulated mechanical stress data, and compares them with preset design threshold data to calculate the physical health component.

[0028] The data-driven model is trained by using LSTM autoencoder model based on normal state data in equipment history and historical operation data. Real-time operation data of SCADA system is input into the model to calculate reconstruction error data, and data health component is calculated based on reconstruction error data. At the same time, survival analysis model is used to predict the remaining lifespan of equipment.

[0029] The knowledge model is driven by matching real-time operating data of the SCADA system, equipment history and historical operating data, and environmental data with rules in the operation and maintenance knowledge graph, triggering inference rules and outputting knowledge health components.

[0030] According to a further improvement of the present invention, the comprehensive health index is generated by fusion, including:

[0031] It receives physical health components, data health components, and knowledge health components.

[0032] The DS evidence theory algorithm is used to perform evidence fusion calculation on the three health components, handle uncertainty conflicts, and generate a comprehensive health index.

[0033] According to a further improvement of the present invention, equipment life prediction and maintenance decision generation includes:

[0034] Historical series data of comprehensive health index are collected, and a time series prediction model is used for learning and prediction to obtain the health index prediction curve and confidence interval data for future time periods.

[0035] When the health index prediction curve reaches the preset warning threshold, a multi-objective optimization model is established with maintenance cost, power outage loss and safety risk as optimization objectives and resources and scheduling as constraints.

[0036] Solve the multi-objective optimization model, output the optimal maintenance time point, the required resource list and operation plan data, and automatically generate structured maintenance work orders.

[0037] According to a further improvement of the present invention, the maintenance strategy is visualized and virtually rehearsed, and augmented reality is used to assist in its execution, including:

[0038] Load work plan data into a parametric 3D model, simulate maintenance operations, and perform physical simulation.

[0039] Calculate the expected operating status data of the equipment after the operation, and verify the solution;

[0040] Maintenance work orders and operation plans are sent to AR glasses worn by on-site personnel. The AR glasses overlay the identification information of the parts to be operated and the animation of the standard operating procedure onto the video screen of the real equipment to guide on-site operation.

[0041] According to a further improvement of the present invention, driving the physical simulation model to calculate the physical health component includes:

[0042] Using real-time operating data and environmental data from the SCADA system as input boundary conditions, the thermodynamic simulation module calculates the internal temperature field distribution data of the equipment and locates overheating risk points.

[0043] The drive electric stress simulation module calculates the electric field distribution data of key insulation components and evaluates the insulation aging rate;

[0044] The mechanical stress simulation module combines load changes and historical short-circuit current data to calculate the cumulative fatigue data of the mechanism components;

[0045] The simulated temperature field distribution data, electric field distribution data, and fatigue accumulation data are compared with the corresponding design thresholds, and the physical health component is calculated by weighted fusion.

[0046] According to a further improvement of the present invention, the data-driven model is used to calculate the data health component, including:

[0047] Preprocess the equipment history and historical operation data, including data cleaning, normalization and time series alignment, to form a standardized historical dataset;

[0048] A baseline model of normal behavior is constructed by training an LSTM autoencoder using a standardized historical dataset.

[0049] Real-time SCADA data is input into the normal behavior baseline model to calculate the reconstruction error at each time point, and anomaly scores are generated based on the reconstruction error sequence.

[0050] By combining the remaining life expectancy data output by the survival analysis model, the health component of the data is calculated through linear combination.

[0051] According to a further improvement of the present invention, evidence fusion calculation is performed using the DS evidence theory algorithm, including:

[0052] Assign basic probability assignment functions to the physical health component, data health component, and knowledge health component respectively;

[0053] Calculate the conflict factors between each health component and adjust the evidence weights based on the conflict factors;

[0054] The adjusted evidence was iteratively fused using the Dempster combination rule to obtain the probability distribution of the comprehensive health index;

[0055] The expected value is extracted from the probability distribution as the final comprehensive health index.

[0056] Beneficial effects: This invention achieves accurate quantification and early warning of the health status of power equipment through a dynamic health assessment mechanism that integrates multiple models, upgrading the operation and maintenance mode from post-maintenance to predictive maintenance; it overcomes the drawbacks of data and model separation, enabling operation and maintenance decisions to shift from relying on experience to data-driven, improving equipment reliability, extending service life, and reducing maintenance costs and power outage risks. Attached Figure Description

[0057] Figure 1 This is a flowchart of the overall steps of an embodiment of the present invention.

[0058] Figure 2 This is a flowchart illustrating the construction of a parameterized digital twin of a power device according to an embodiment of the present invention.

[0059] Figure 3 This is a flowchart of the health component of the multi-model collaborative computing device according to an embodiment of the present invention. Detailed Implementation

[0060] like Figure 1 As shown, the methods for dynamic health assessment and predictive maintenance of power equipment include:

[0061] S1: Constructing a parametric digital twin of power equipment

[0062] S1.1: Collect basic attribute data of the equipment, perform parametric modeling, and obtain a parametric 3D model.

[0063] The system collects design drawings, material properties, and rated parameters of the equipment; it then calls a parametric modeling engine to construct a parametric 3D model with dynamically adjustable key dimensions and properties based on the design drawings and material properties.

[0064] S1.2: Integrates a multiphysics simulation module to build a physical simulation model.

[0065] Configure thermodynamic simulation parameters, electrical stress simulation parameters, and mechanical stress simulation parameters; associate the parameterized three-dimensional model with the multiphysics simulation parameters to construct a physical simulation model that can perform dynamic calculations based on input boundary conditions.

[0066] S1.3: Construct an operations and maintenance knowledge graph to form a knowledge model.

[0067] Collect equipment technical manuals, operation and maintenance procedures, historical fault records and expert experience rules; perform natural language processing and rule extraction to construct an operation and maintenance knowledge graph with equipment components, fault modes and operation and maintenance actions as entities and causal relationships and conditional triggering relationships as edges.

[0068] S2: Real-time data acquisition and injection

[0069] S2.1: Collect real-time operating data of the SCADA system.

[0070] The system periodically collects device current, voltage, temperature, and switch status data through a data interface.

[0071] S2.2: Collect equipment history and historical operation data.

[0072] From the equipment history database, query and collect maintenance record data, historical load data and past fault data of the target equipment.

[0073] S2.3: Collect environmental and inspection data.

[0074] It collects ambient temperature and humidity data reported by sensors deployed on-site, as well as inspection images and recorded data uploaded by inspection personnel via mobile terminals.

[0075] S3: Multi-model collaborative computing device health component

[0076] S3.1: Drive the physical simulation model and calculate the physical health components.

[0077] The real-time operating data of the SCADA system and the environmental data are used as boundary conditions and input into the physical simulation model; multiphysics simulation calculations are performed to obtain the simulated temperature field distribution data, simulated electrical stress distribution data and simulated mechanical stress data of the equipment; the simulation data are compared with the preset design threshold data to calculate the physical health component.

[0078] S3.2: Drive the data-driven model and calculate the data health component.

[0079] Based on the normal state data from the equipment's history and historical operation data, an LSTM autoencoder model is trained to obtain a normal behavior baseline model. The real-time operation data of the SCADA system is input into this normal behavior baseline model, and reconstruction error data is calculated and output. Based on the magnitude of the reconstruction error data, a data health component is calculated. Simultaneously, a survival analysis model is trained using historical data to predict the remaining lifespan of the equipment.

[0080] S3.3: Drive the knowledge model and calculate the knowledge health component.

[0081] The real-time operation data, equipment history and historical operation data, and environmental data of the SCADA system are matched with the rules in the operation and maintenance knowledge graph; the inference rules that meet the conditions are triggered, and an experience-based health status score, i.e., the knowledge health component, is output.

[0082] S4: Integrates outputs from multiple models to calculate a comprehensive health index.

[0083] The system receives the physical health component, data health component, and knowledge health component; it then uses the DS evidence theory algorithm to fuse and calculate the evidence carried by the three health components, handle their uncertainty conflicts, and obtain a comprehensive health index that uniquely represents the overall health status of the device.

[0084] S5: Perform lifespan prediction and maintenance decision generation

[0085] S5.1: Predict device health trends based on time series data.

[0086] Historical sequence data of the comprehensive health index are collected; a time series prediction model is used to learn and predict the historical sequence data to obtain the health index prediction curve and confidence interval data for future time periods.

[0087] S5.2: Generate the optimal maintenance strategy.

[0088] When the health index prediction curve reaches the preset warning threshold, the decision-making process is triggered; a multi-objective optimization model is established with maintenance cost, power outage loss and safety risk as optimization objectives and resources and scheduling as constraints; the multi-objective optimization model is solved, the optimal maintenance time point, the required resource list and work plan data are output, and a structured maintenance work order is automatically generated.

[0089] S6: Visual simulation and on-site execution of maintenance plans

[0090] S6.1: Conduct virtual debugging and rehearsal of the maintenance plan.

[0091] The operation plan data is loaded into the parametric 3D model; maintenance operations are simulated and executed; and the expected operating status data of the equipment after the operation is calculated based on the physical simulation model, so that maintenance personnel can verify the plan.

[0092] S6.2: Assist in on-site maintenance execution using AR devices.

[0093] The maintenance work order and operation plan data are sent to the AR glasses of on-site personnel. In the AR glasses, the identification information of the parts to be operated and the animation of the standard operation process are overlaid on the real equipment video screen to guide the on-site personnel to operate.

[0094] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for dynamic health assessment and predictive maintenance of power equipment, characterized in that, include: Construct a parameterized digital twin of power equipment and collect real-time operation data, historical data and environmental data; Based on the parameterized digital twin and the collected data, the device health component is calculated using a multi-model collaborative method and then integrated to generate a comprehensive health index. Based on the comprehensive health index, equipment lifespan is predicted and maintenance decisions are generated, and the optimal maintenance strategy is output. The optimal maintenance strategy is visualized and virtually simulated, and then executed with augmented reality assistance to form a closed-loop maintenance system.

2. The method according to claim 1, characterized in that, Constructing parameterized digital twins of power equipment includes: Collect equipment design drawings, material property data, and rated parameter data, perform parametric modeling, and obtain a parametric 3D model; It integrates a multiphysics simulation module, configures thermodynamic simulation parameters, electrical stress simulation parameters, and mechanical stress simulation parameters, and constructs a physical simulation model; Collect equipment technical manuals, operation and maintenance procedures, historical fault records, and expert experience rules, perform natural language processing and rule extraction, construct an operation and maintenance knowledge graph, and form a knowledge model.

3. The method according to claim 1, characterized in that, Collect real-time operational data, historical data, and environmental data, including: The system collects device current data, voltage data, temperature data, and switch status data through a data interface. Query and collect maintenance record data, historical load data, and past fault data of the target equipment from the equipment history database; The system collects ambient temperature and humidity data reported by sensors deployed on-site, as well as inspection images and recorded data uploaded by inspection personnel via mobile terminals.

4. The method according to claim 1, characterized in that, The device health components are calculated using a multi-model collaborative approach, including: The driving physical simulation model uses real-time operating data of the SCADA system and environmental data as boundary conditions to perform multi-physics simulation calculations, obtain simulated temperature field distribution data, simulated electric stress distribution data and simulated mechanical stress data, and compares them with preset design threshold data to calculate the physical health component. The data-driven model is trained by using LSTM autoencoder model based on normal state data in equipment history and historical operation data. Real-time operation data of SCADA system is input into the model to calculate reconstruction error data, and data health component is calculated based on reconstruction error data. At the same time, survival analysis model is used to predict the remaining lifespan of equipment. The knowledge model is driven by matching real-time operating data of the SCADA system, equipment history and historical operating data, and environmental data with rules in the operation and maintenance knowledge graph, triggering inference rules and outputting knowledge health components.

5. The method according to claim 1, characterized in that, The system integrates and generates a comprehensive health index, including: It receives physical health components, data health components, and knowledge health components. The DS evidence theory algorithm is used to perform evidence fusion calculation on the three health components, handle uncertainty conflicts, and generate a comprehensive health index.

6. The method according to claim 1, characterized in that, Perform equipment life prediction and maintenance decision generation, including: Historical series data of comprehensive health index are collected, and a time series prediction model is used for learning and prediction to obtain the health index prediction curve and confidence interval data for future time periods. When the health index prediction curve reaches the preset warning threshold, a multi-objective optimization model is established with maintenance cost, power outage loss and safety risk as optimization objectives and resources and scheduling as constraints. Solve the multi-objective optimization model, output the optimal maintenance time point, the required resource list and operation plan data, and automatically generate structured maintenance work orders.

7. The method according to claim 1, characterized in that, Visualized virtual rehearsals and augmented reality-assisted execution of maintenance strategies include: Load work plan data into a parametric 3D model, simulate maintenance operations, and perform physical simulation. Calculate the expected operating status data of the equipment after the operation, and verify the solution; Maintenance work orders and operation plans are sent to AR glasses worn by on-site personnel. The AR glasses overlay the identification information of the parts to be operated and the animation of the standard operating procedure onto the video screen of the real equipment to guide on-site operation.

8. The method according to claim 4, characterized in that, The physical simulation model is driven to calculate the physical health components, including: Using real-time operating data and environmental data from the SCADA system as input boundary conditions, the thermodynamic simulation module calculates the internal temperature field distribution data of the equipment and locates overheating risk points. The drive electric stress simulation module calculates the electric field distribution data of key insulation components and evaluates the insulation aging rate; The mechanical stress simulation module combines load changes and historical short-circuit current data to calculate the cumulative fatigue data of the mechanism components; The simulated temperature field distribution data, electric field distribution data, and fatigue accumulation data are compared with the corresponding design thresholds, and the physical health component is calculated by weighted fusion.

9. The method according to claim 4, characterized in that, The data-driven model calculates data health components, including: Preprocess the equipment history and historical operation data, including data cleaning, normalization and time series alignment, to form a standardized historical dataset; A baseline model of normal behavior is constructed by training an LSTM autoencoder using a standardized historical dataset. Real-time SCADA data is input into the normal behavior baseline model to calculate the reconstruction error at each time point, and anomaly scores are generated based on the reconstruction error sequence. By combining the remaining life expectancy data output by the survival analysis model, the health component of the data is calculated through linear combination.

10. The method according to claim 5, characterized in that, Evidence fusion calculations are performed using the DS evidence theory algorithm, including: Assign basic probability assignment functions to the physical health component, data health component, and knowledge health component respectively; Calculate the conflict factors between each health component and adjust the evidence weights based on the conflict factors; The adjusted evidence was iteratively fused using the Dempster combination rule to obtain the probability distribution of the comprehensive health index; The expected value is extracted from the probability distribution as the final comprehensive health index.

Images