PHM model training data generation method and device based on digital twinning

Abstract

The invention discloses a PHM model training data generation method and device based on digital twinning, and belongs to the technical field of aircrafts. The method comprises the steps that a medium-fidelity digital twinborn model is constructed, and the model comprehensively considers dynamic characteristics and coupling relations of all components of an aviation system so as to reproduce global dynamic behaviors; the medium-fidelity digital twin model is calibrated based on expert knowledge and historical data so as to improve the accuracy; model simulation is realized in a Monte Carlo batch simulation mode, and labels are added to data according to simulation working conditions so as to generate a large amount of PHM model training data. According to the method, global dynamic behaviors under the coupling influence of all parts of the aviation system can be reflected, large-batch simulation tests can be rapidly executed, training data covering multiple fault modes are generated, and balance between simulation precision and efficiency is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of aircraft technology, and specifically relates to a method and device for generating PHM model training data based on digital twins. Background Art

[0002] Prognostic health management (PHM) is a key technical approach to improving the reliability and safety of aviation equipment. Data-driven PHM models typically rely on extensive training with extensive historical operational and fault data to accurately predict and diagnose system health. However, due to the high reliability of aviation systems, the frequency of actual failures is extremely low, making it difficult to obtain sufficient real-world failure data. Even deliberately creating failures through testing is not only costly and poses safety risks, but the resulting sample size is still limited, making it difficult to fully cover all failure modes. This data shortage is particularly pronounced in the aviation sector, becoming a bottleneck restricting the accuracy and practical application of PHM models.

[0003] To address the lack of real-world data, existing techniques have attempted to use simulation to generate synthetic fault data for model training. However, traditional simulation methods suffer from two limitations: First, while low-precision or localized, simple models are computationally fast, they fail to accurately reflect the fault evolution process under the coupled interactions of various components in complex aviation systems, resulting in low reliability of the simulated data. Second, while simulations based on high-precision physical models offer realistic results, they are extremely expensive to build and run, making them difficult to use for generating large datasets. Digital twin technology offers a new approach to this problem: creating models that closely align with the behavior of the real system in a virtual environment, enabling various virtual experiments to be conducted. However, for complex systems like aviation, pursuing a completely high-fidelity digital twin model faces the challenges of complex model construction and significant computational resource consumption, making it difficult to efficiently generate massive amounts of data. Conversely, oversimplified models can make it difficult to ensure that simulation results are effectively representative of real-world faults. Therefore, a method that strikes a balance between simulation accuracy and efficiency is urgently needed. Furthermore, existing techniques fail to adequately consider the scenario of compound faults (i.e., multiple faults occurring simultaneously), where data is even more scarce. This results in poor performance of PHM models in multi-fault scenarios. Summary of the Invention

[0004] The present invention aims to solve one of the technical problems in the above-mentioned related art at least to a certain extent.

[0005] To this end, the purpose of the present invention is to provide a method and device for generating PHM model training data based on digital twins, which can achieve a balance between simulation accuracy and efficiency.

[0006] In order to solve the above-mentioned technical problems, the present invention is achieved as follows:

[0007] An embodiment of the present invention provides a method for generating PHM model training data based on digital twins, the method comprising:

[0008] Build a medium-fidelity digital twin model that comprehensively considers the dynamic characteristics and coupling relationships of each component of the aviation system to reproduce the global dynamic behavior;

[0009] The medium-fidelity digital twin model is calibrated based on expert knowledge and historical data to improve accuracy;

[0010] Model simulation is achieved through Monte Carlo batch simulation, and data are labeled according to simulation conditions to generate a large amount of PHM model training data.

[0011] In addition, the method for generating PHM model training data based on digital twins according to the present invention may also have the following additional technical features:

[0012] In some embodiments, the PHM model training data is normal data and fault data with state labels, and the fault data includes single fault data and compound fault data.

[0013] In some embodiments, the Monte Carlo batch simulation performs multiple simulation tests by randomly changing the initial state, operating conditions, and failure mode parameters of the aviation system, wherein the failure mode parameters include the failure type, occurrence time, and severity; and injecting transient / progressive failures and compound failures.

[0014] In some implementations, the composite fault is two or more different types of faults injected simultaneously into a simulation test.

[0015] In some embodiments, the medium-fidelity digital twin model includes an engine propulsion sub-model, a flight control sub-model, an avionics and sensor sub-model, and an airframe structural dynamics sub-model.

[0016] In some embodiments, the medium-fidelity digital twin model retains subsystem interactions critical to PHM, including:

[0017] Flight-engine coupling: the effects of aircraft attitude and speed on engine intake and loads, and the interaction of engine thrust reactions to flight conditions;

[0018] Coupling of power generation and consumption: the impact of engine speed on generator output, and the impact of electrical faults on flight control and instrumentation;

[0019] The impact of structural degradation on performance: increased drag of the aircraft and the impact of structural deformation on handling stability.

[0020] In some embodiments, the effect of structural degradation on performance is reflected in medium fidelity using parameter perturbations, including reducing the slope of the lift curve by introducing a health factor and characterizing the effect of structural aging on flight performance by increasing the drag coefficient.

[0021] In some embodiments, the dynamic characteristics of each component include:

[0022] Flight dynamics: uses an open source 6-degree-of-freedom nonlinear engine to simulate attitude and trajectory;

[0023] Engine and propulsion systems: Modeling thermodynamics, rotor dynamics, and thrust / fuel consumption using Modelica;

[0024] Power supply and other systems: Use Modelica to build models of the gas source and hydraulic subsystems to capture key physical evolutions.

[0025] In some embodiments, the medium-fidelity digital twin model simplifies non-essential details, including:

[0026] Detailed blade stress distribution inside the engine;

[0027] The electrical system is simulated using an equivalent gas circuit model;

[0028] The aerodynamics of the aircraft configuration uses lookup table data.

[0029] An embodiment of the present invention also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the content of the method for generating PHM model training data based on digital twins as described in any one of the above items.

[0030] Compared with the prior art, the present invention has at least the following beneficial effects:

[0031] In the embodiments of the present invention, the provided method for generating PHM model training data based on digital twins can not only reflect the global dynamic behavior under the influence of the coupling of various components of the aviation system, but also quickly perform large-scale simulation tests to generate training data covering various failure modes;

[0032] The digital twin-based PHM model training data generation method provided in the embodiments of the present invention can obtain a large amount of normal and fault data with status labels in a virtual environment, allowing for dangerous and expensive physical fault testing, significantly alleviating the problem of a lack of fault samples in PHM model training.

[0033] In the embodiments of the present invention, the digital twin-based PHM model training data generation method provided adopts a medium-fidelity modeling approach in terms of simulation efficiency and accuracy, fully considering the coupling effects between various components of the aviation system. While ensuring a certain level of simulation accuracy, it simplifies the model complexity, greatly improving the simulation operation efficiency and supporting more than a thousand Monte Carlo batch simulations to generate massive amounts of data.

[0034] In the embodiments of the present invention, the provided method for generating PHM model training data based on digital twins introduces expert experience and real historical data to calibrate the parameters of the digital twin model in terms of model credibility. This enables the model to accurately reproduce the dynamic behavior characteristics and typical fault symptoms of the actual system, ensuring that the generated data has high credibility and engineering reference value.

[0035] In the embodiments of the present invention, the provided method for generating training data for a PHM model based on digital twins has a richness of fault scenarios. The simulation test covers a variety of fault modes, including sensor failure, actuator failure, component performance degradation, etc. It can even simulate the occurrence of multiple different faults simultaneously in a single simulation. Therefore, the generated data set covers single fault and compound fault scenarios, which helps to train a PHM model that can identify complex fault combinations and improve the robustness and accuracy of fault diagnosis and prediction.

[0036] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a diagram of the overall architecture of a method for generating PHM model training data based on digital twins, disclosed in one embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0039] The embodiments of the present invention are described in detail below through specific embodiments and application scenarios with reference to the accompanying drawings.

[0040] In some embodiments of the present invention, a method for generating PHM model training data based on a medium-fidelity aviation system digital twin model is provided, which includes:

[0041] First, a medium-fidelity digital twin model of the aviation system is constructed, recreating the key components of the actual aviation system, their interconnected relationships, and their global dynamic behavior in a virtual environment. This digital twin model is calibrated and refined by incorporating expert knowledge and historical data, enabling it to realistically simulate the operating characteristics of the aircraft under normal and fault conditions.

[0042] Subsequently, large-scale simulation experiments were conducted on the digital twin model using Monte Carlo simulation methods. This involves repeated simulations with varying initial conditions and operating parameters. Some of these simulations were conducted under normal conditions, while others involved various fault modes (including single faults and combined faults) according to pre-defined requirements. This generated a wealth of data on both normal and faulty behavior.

[0043] Finally, the acquired simulation data is labeled with corresponding status labels (normal or specific failure mode) to form a large-scale, multi-label PHM model training dataset. The present invention also provides a digital twin system for implementing the above method, comprising a model construction module, a simulation execution module, and a data processing module, which completes the entire process of establishing and calibrating the digital twin model and generating Monte Carlo simulation data.

[0044] See also Figure 1 As shown, in some embodiments of the present invention, the digital twin system used for PHM training data generation includes core components such as a digital twin model module, a simulation execution module, and a data processing module. The digital twin model module is used to build a virtual model of the aviation system, simulating the operating characteristics of the aircraft's major components and their interactions. The simulation execution module is used to control the operation and test configuration of the digital twin model, implementing Monte Carlo batch simulation. The data processing module is used to collect, process, and annotate simulation output data.

[0045] Preferably, the system is also provided with an expert knowledge base and a historical data interface for storing the experience rules of domain experts, failure mode information, and historical data obtained from actual flights or tests, and providing this information to the digital twin model module and the simulation execution module to calibrate the model and guide the simulation test settings so that the model can realistically simulate the normal state and failure state of the aviation system.

[0046] In some implementations of the present invention, the digital twin model uses the Monte Carlo method to perform multiple simulations. Some simulations are conducted under normal operating conditions, while others introduce pre-defined failure modes, including single failure scenarios and complex failure scenarios involving multiple failures. The operational data generated by each simulation is collected and labeled as normal or with the corresponding failure mode based on the simulated operating conditions. This data is then aggregated to form a training dataset for the PHM model.

[0047] First, a digital twin model of the aviation system is established. This model offers medium-fidelity simulation accuracy and comprehensively considers the dynamic characteristics of the aircraft's major subsystems and their coupling relationships. For example, in one embodiment, the digital twin model includes engine propulsion submodels, flight control submodels, avionics and sensor submodels, and airframe structural dynamics submodels. These submodels are coupled through internal interfaces and equations to jointly simulate the overall flight dynamics of the aircraft. Each component submodel can be built based on known physical principles and calibrated with real-world data. For example, historical flight test data is compared with simulation outputs, and the parameters of the engine thrust model are continuously adjusted to ensure that the output under normal operating conditions is consistent with the actual performance curve. Alternatively, certain elastic parameters in the structural model can be corrected based on empirical formulas provided by experts to more accurately reflect the vibration characteristics of the airframe. Through these calibration measures, the digital twin model achieves a high degree of consistency with the real system in key performance indicators, laying the foundation for subsequent simulations.

[0048] Monte Carlo simulations are then conducted based on the calibrated digital twin model. Specifically, the simulation execution module sets the total number of simulations and parameter ranges. At the beginning of each independent simulation, a random combination of initial conditions and input parameters is selected. For normal operating simulations, the system randomly varies parameters such as environmental conditions (e.g., ambient temperature, airspeed), mission profile (e.g., payload, trajectory), and component performance deviations (e.g., sensor noise, manufacturing tolerances) within a safe range. Through repeated simulations, data samples covering various normal operating conditions are obtained. For fault simulations, while randomizing normal parameters, fault events are introduced into the simulation according to pre-set probabilities or test plans. Faults can be triggered at random moments in the simulation timeline or set to exist from the start of the simulation, simulating transient or progressive degradation faults as needed. For example, in one simulation, a sensor output can be caused to drift at a random moment mid-flight to simulate sensor misalignment; or in another simulation, the maximum engine thrust can be gradually reduced by a certain amount to simulate engine performance degradation. For complex fault scenarios, the simulation execution module can simultaneously inject two or more different fault types. For example, a hydraulic system leak and a controller sensor failure can be simultaneously applied in the same simulation to examine their combined impact on the aircraft's dynamic performance. By systematically and randomly varying various factors and combining different fault modes, the present invention can generate simulation data covering a wide range of operating conditions.

[0049] During each simulation, the digital twin model outputs a large amount of data representing system status and performance, such as sensor readings, operating parameters of key components (temperature, pressure, current, etc.), and flight status parameters (speed, altitude, attitude). The data processing module collects and organizes this raw simulation output data in real time. At the end of the simulation, the operating status category is recorded based on the simulation settings: if no faults were introduced, the data is labeled "normal"; if a certain fault was included in the simulation, the corresponding fault type is assigned; if multiple faults were included, multiple labels or composite labels are created for the combined fault mode. The data processing module associates the labels with the simulated data sequences and stores them as samples for training the PHM model. After numerous simulations, the system ultimately aggregates a large-scale dataset containing both normal samples and samples of various fault modes. This dataset can be used to train and validate data-driven PHM algorithm models. For example, the data generated by the present invention can be fed into a machine learning model (such as a neural network or random forest) for training, enabling it to learn the differential characteristics of sensor data under normal and fault conditions, thereby enabling automatic fault detection and identification during deployment. For example, the accuracy of remaining useful life (RUL) prediction can be improved by training a prediction model with time series data that includes the component performance degradation process.

[0050] It should be noted that the construction and simulation of the digital twin model of the present invention can be achieved using existing mature modeling and simulation platforms, such as flight mechanics simulation software, MATLAB / Simulink modeling environment, etc. It is only necessary to build the corresponding modular model according to the idea of ​​the present invention and write the fault injection and data acquisition program. Therefore, the method and system of the present invention have strong versatility and can be promoted and applied to the generation of PHM data for various types of aircraft (such as airplanes, drones, etc.). In addition, without departing from the principles of the present invention, the specific component division, parameter selection, fault type setting, etc. in the above embodiments can be adjusted and transformed according to different application requirements, and the protection scope of the present invention is not limited to a specific aircraft or a specific fault type. Through the description of the above embodiments, those skilled in the art can clearly understand the method flow and system architecture of the present invention, and implement and apply the present invention accordingly.

[0051] Example 1:

[0052] This implementation builds a digital twin simulation platform that supports evolutionary-based PHM (Prognostics and Health Management) model training and discovery in civil aircraft scenarios. The platform emphasizes reproducibility and engineering feasibility, integrating the dynamics and maintenance processes of aircraft subsystems to provide high-quality simulation data for training and evaluating PHM algorithms in data-sparse scenarios. Key objectives include:

[0053] 1) Continuous + discrete hybrid simulation: Unified simulation of aircraft continuous physical processes (flight dynamics, engine propulsion, electrical power supply, etc.) and discrete events (maintenance and repair, fault occurrence, scheduling decisions, etc.) to form a system-level hybrid simulation environment.

[0054] 2) Medium-fidelity modeling: Use medium-fidelity physical models, focusing on key coupling mechanisms and global dynamic behaviors rather than overly detailed CAE models, to strike a balance between accuracy and computational overhead.

[0055] 3) Uncertainty and Monte Carlo: Supports modeling of multi-source uncertainty, such as environmental disturbances and component differences, and can perform batch runs of Monte Carlo simulations for multiple scenarios, generating large amounts of diverse data to compensate for the lack of real data.

[0056] 4) Fault injection and health evolution: Typical faults (engine thrust loss, electrical power failure, structural fatigue degradation, etc.) can be injected into the simulation to track the performance degradation and remaining useful life (RUL) evolution of each component.

[0057] 5) Evolutionary PHM Training Integration: Tightly integrated with the evolutionary PHM model development process, it enables automatic training of model populations, fitness evaluation and selection based on metrics such as accuracy and "belief-reliability," and automatic iterative optimization of models.

[0058] The following details the platform's module design, architecture integration, tool chain selection, and implementation details to ensure that researchers can reproduce and conduct secondary development based on this solution.

[0059] The first aspect is the fusion simulation of continuous physical processes and discrete maintenance events.

[0060] Continuous dynamics simulation: Use physical modeling to simulate the aircraft's flight and system continuous processes. The core includes:

[0061] a. Flight Dynamics: Use 6-DOF flight dynamics models such as JSBSim to simulate aircraft attitude, trajectory, and flight state. JSBSim is an open-source, nonlinear 6-DOF flight dynamics engine that is data-driven and suitable for batch simulation.

[0062] b. Engine and Propulsion Systems: Modelica is used to model the thermodynamic and rotordynamic behavior of aircraft engines, as well as thrust output and fuel consumption. As a multi-domain modeling language, Modelica facilitates the creation of one-dimensional, moderately complex models of engines and transmissions, capturing coupled relationships such as speed, thrust, and temperature.

[0063] c. Power and other physical systems: Modelica can also be used to build electrical system models (generator, APU, battery, etc.), as well as subsystem models such as hydraulics and avionics cooling to include the physical evolution of critical aircraft systems.

[0064] Discrete Events and Maintenance Strategies: Introducing maintenance-related discrete behaviors through discrete event simulation (DES) or agent / rule-driven models. The core includes:

[0065] a. Failure and Maintenance Events: Simulate random or triggered failure events, as well as regular maintenance activities such as inspection, repair, and component replacement. Use an event scheduling mechanism to schedule these discrete events on a timeline and update the system state when an event occurs (e.g., resetting damage accumulation after replacing a component).

[0066] b. Intelligent maintenance / scheduling strategies: Decision algorithms or rules (implemented by agents or policy functions) are introduced to trigger maintenance decisions based on system health and forecast information. For example, preventive maintenance can be scheduled when the predicted RUL falls below a threshold, or overall availability can be optimized based on multi-fleet scheduling. The simulation platform allows policy algorithms to be embedded and their impact on performance and availability to be observed during operation.

[0067] c. Tool Implementation: We recommend using multi-method simulation software such as AnyLogic to implement the aforementioned discrete component models. AnyLogic supports hybrid modeling using diverse methodologies, including discrete events, agents, and system dynamics. This facilitates integrating maintenance processes (personnel, resources, and task flows) with the continuous evolution of system states. For example, a flowchart can be used to describe a process such as inspection-repair-release, and linked to the continuous degradation process. For open-source solutions, Python's SimPy library or our proprietary event scheduling module can also be used to implement similar functionality.

[0068] Hybrid simulation integration: enables the coordinated operation of continuous and discrete parts. The core includes:

[0069] a. Use a hybrid scheduling algorithm combining time-stepping and event-driven scheduling: Continuous subsystems (such as flight and engine systems) are typically integrated with fixed, small steps (e.g., 0.01–0.1 seconds). Discrete events are sorted by their trigger time in the scheduling queue. The recommended hybrid simulation framework checks the next event time within each simulation loop. If no event occurs, the continuous simulation proceeds. If an event is imminent, the simulation jumps to the event time, processes the event logic, and then continues.

[0070] b. Data Exchange: Discrete events (failures) can modify the parameters / state of continuous models. For example, an engine failure event can cause a decrease in engine thrust (achievable by adjusting the internal efficiency parameters of the engine model). Conversely, the state of continuous models can also influence discrete decisions, such as when the real-time calculated RUL triggers a maintenance event. A unified data interface is required to allow event modules to access and modify key states / parameters of the physical model. A publish-subscribe or global state management mechanism can be used to centrally manage the state variables of all aircraft subsystems within the simulation. After each event or time step, the state is written to central storage and then made available to other modules.

[0071] c. Verify synchronization: Ensure that continuous integration and event processing are strictly synchronized with the simulation clock. For example, use synchronization barriers at each event occurrence to prevent the continuous simulation from overshooting the event time. Tools like AnyLogic already natively support synchronization of event scheduling and continuous simulation in hybrid modeling. In custom implementations, you can advance through time in a Python loop and actively check the event list.

[0072] Through the above design, unified simulation of aircraft operation (continuous dynamics) and maintenance operation (discrete events) is achieved, fully reproducing the actual operating conditions.

[0073] Part II, medium-fidelity modeling with key coupling preservation.

[0074] Definition of Medium Fidelity: Medium-fidelity models fall between simple, low-fidelity empirical models and complex, high-fidelity CAE models. Their goal is to accurately reproduce the critical system behaviors while avoiding excessive computational overhead. According to Emerson's Digital Twin Simulation Guide, medium-fidelity models typically use first-order principles physics models with appropriate simplifications to capture mass, energy conservation, and key dynamic characteristics. Compared to purely data-driven models, medium-fidelity models offer more reliable extrapolation to out-of-scope conditions.

[0075] Key coupling mechanisms: To ensure that PHM-related behaviors are simulated, the model needs to retain key couplings between subsystems and global dynamics. The core includes:

[0076] a. Flight-Engine Coupling: Aircraft attitude and speed affect engine intake and load, and conversely, engine thrust influences aircraft motion. This interaction requires modeling. The solution is to exchange parameters between the JSBSim flight model and the engine Modelica model. For example, current airspeed and air density are passed to the engine model to calculate pressure ratio and thrust, and the thrust and torque generated by the engine are fed back to the flight model to calculate acceleration.

[0077] b. Coupling power generation and consumption: Engine speed affects generator output, and electrical faults can affect flight control systems and instrumentation. Power and speed parameters must be exchanged between the engine and electrical models. For example, the Modelica electrical system submodel uses engine speed and torque as input to calculate generated power, output bus voltage to supply other systems, and simulate battery charging and discharging.

[0078] c. Structural degradation impacts performance: For example, increased drag from the airframe and structural deformation affecting control stability. This can be achieved with medium fidelity using parameter perturbations. By introducing a "health factor" to reduce the slope of the lift curve and increase the drag coefficient, the impact of structural aging on flight performance can be simply characterized. This approach, even without detailed finite element analysis, can capture the impact of degradation on global dynamics.

[0079] Simplification of non-essential details: discarding minor details that have no significant impact on PHM training. For example:

[0080] a. Detailed blade stress distribution inside the engine does not require CAE modeling, and damage progression can be represented by an overall efficiency and flow degradation coefficient.

[0081] b. The electrical system does not require phase-by-phase electromagnetic simulation, but instead uses an equivalent circuit model (air pressure source + internal resistance, etc.).

[0082] c. Aircraft configuration aerodynamics can be analyzed using table lookup data (Wind Tunnel database or JSBSim's own XML) rather than real-time CFD.

[0083] By focusing on key couplings and degradation effects and simplifying minor details, the model improves simulation speed while ensuring predictive relevance, making it suitable for conducting large numbers of Monte Carlo experiments.

[0084] Part III, Multi-source uncertainty modeling and Monte Carlo data generation.

[0085] Sources of uncertainty: The platform models multiple sources of uncertainty through parameter randomization and disturbance injection, including but not limited to:

[0086] a. Environmental uncertainty: Random changes in the external environment, such as atmospheric temperature, air pressure, wind field disturbances, and turbulence intensity, can be introduced into the atmospheric model during simulation. For example, encountering turbulent gusts of varying intensities or temperature deviations can affect the engine and aerodynamics.

[0087] b. Individual differences: Manufacturing and wear variations between different aircraft and components. Initial health parameters can be randomly sampled, such as a slightly lower initial efficiency for a particular engine or different initial damage factors for certain structures.

[0088] c. Operational and Load Variation: Pilot operations and mission profiles vary, resulting in slightly different payloads, routes, and flight profiles for each flight, affecting component stress spectra. Different mission scenarios (including climb / cruise / descent times, altitudes, and power settings) can be extracted from the simulation.

[0089] d. Measurement and model uncertainty: Sensor noise, model simplification errors, etc. can also be reflected by superimposing noise on the output signal.

[0090] Monte Carlo scenario generation: supports batch automated simulation and runs of a large number of scenarios, achieving data sampling coverage; core features include:

[0091] a. Parameter combination sampling: Monte Carlo random sampling or Latin hypercube methods are used to perform multiple sampling of the aforementioned uncertain parameters, generating hundreds or even thousands of sets of scenario parameters. Each set of parameters defines a possible aircraft, environment, and usage scenario.

[0092] b. Parallel simulation: Distribute simulation tasks using Python parallel libraries such as Ray. Ray executes each scenario simulation as a remote task, leveraging multi-core or cluster parallel acceleration. This allows for large-scale sample generation within an acceptable timeframe.

[0093] c. Automatic Data Logging: Each simulation run automatically outputs the required data (sensor timing, fault occurrence time, RUL curves, maintenance action records, etc.). The platform stores this data in a unified format, such as a time-series CSV or SQL database (see the PHM Society UAV framework for support for automatic simulation data logging and verification). Ensure that metadata (such as parameter seeds and fault types) is bound to the results to ensure traceability of the training algorithm.

[0094] By using Monte Carlo simulations to simulate a large number of possible scenarios, PHM model training can obtain a rich and diverse dataset, including degradation processes under different failure modes, different initial health states, and different environmental stresses. This is crucial for improving algorithm robustness and generalization capabilities in data-sparse scenarios.

[0095] Part 4, fault injection mechanism and health status / RUL tracking.

[0096] Typical Failure Mode Injection: The platform will have built-in simulation injection mechanisms for a variety of representative aviation system failures; the core includes:

[0097] a. Engine Thrust Loss / Degradation: Thrust loss is simulated by adjusting engine model parameters. For example, engine performance degradation is simulated by gradually decreasing the compressor and combustion efficiency curves, resulting in a decrease in maximum available thrust. A sudden drop in efficiency may represent a sudden failure (such as compressor stall or turbine damage), while a gradual, smaller drop indicates wear and tear. NASA's C-MAPSS model uses exponential decay of flow and efficiency parameters to generate degradation data.

[0098] b. Electrical system power outage: This simulates generator failures and busbar power outages. At a specific point in the simulation, the output of a generator in the electrical network can be triggered to zero, switching the load to the backup system, causing a brief voltage drop. Alternatively, a rapid battery discharge failure can cause a sustained drop in bus voltage, requiring emergency power supply intervention.

[0099] c. Structural / Component Fatigue Degradation: A fatigue life model (such as the Paris crack growth formula or a simple life deduction model) is configured for structural components. Damage is accumulated based on the stress spectrum at each time step, and component failure is marked when the damage exceeds a threshold. A performance degradation event can also be triggered directly at a specific point in the simulation for a component (such as a hydraulic pump) to indicate critical wear.

[0100] Therefore, the health index of the component decreases over time in the simulation.

[0101] d. Sensor or control failure: Although the focus is on predicting RUL, it is also possible to simulate sensor misalignment and controller failure to enrich the training scenarios of the PHM algorithm (these are more complex failures and can be included optionally).

[0102] RUL and health indicator tracking: Establish health status variables for each key component and calculate RUL; the core includes:

[0103] a. Health Index (HI): Defines a health index ranging from 0 to 1 or a percentage, where 1 represents a new, intact component and 0 indicates failure. During simulation, the HI is updated based on the degradation model. For example, the HI of an engine compressor blade can be calculated based on the efficiency reduction rate (an efficiency decrease of X% corresponds to a HI decrease of Y%). The HI of a structural component is calculated based on the accumulated fatigue cycles consumed.

[0104] b. Remaining Useful Life (RUL): When health indicator trends are predictable, the RUL is the estimated remaining operational time (or number of cycles) at the current moment. In simulation, the true RUL can be directly calculated using a known degradation model. For example, if a component is set to fail when its HI drops to 0 in a simulation, the true RUL can be calculated based on the current HI value and the rate of decline. For random failures, a life limit can be randomly drawn at the start of the simulation and then counted down during operation.

[0105] c. Performance evolution records: In addition to HI and RUL, corresponding performance indicators must also be tracked, such as engine thrust and fuel consumption degradation curves, and battery capacity decay curves over cycles. This not only verifies the PHM model's ability to predict performance degradation, but also enables decision-making in subsequent maintenance strategy simulations (e.g., if performance drops below a threshold, maintenance is required).

[0106] d. Data Interface: The platform records the HI and RUL of all components as special outputs over time and provides an interface to the PHM algorithm. This means that the PHM model can use simulated HI / RUL as labels during offline training or compare the simulated RUL with the actual RUL to evaluate accuracy during online testing.

[0107] Through fault injection and health tracking mechanisms, simulation platforms can generate operational data with fault and lifespan labels, meeting the requirements for degradation process and lifespan information in PHM algorithm development. As Saxena et al. pointed out, obtaining complete data on a real system from health to failure is extremely difficult, but simulation injection provides a feasible way to generate such data for model training and validation.

[0108] Part 5, PHM training closed-loop integration based on evolutionary mechanism.

[0109] Evolutionary PHM model training process: The platform supports embedding the simulation environment into the automated training and optimization loop of the PHM model, forming a closed loop of digital twin + evolutionary algorithm:

[0110] a. Model population initialization: First, a set of candidate PHM models are generated (which can be neural networks with different structures, random forests, or models with different hyperparameters). These models, as individuals in the population, will be iteratively optimized through an evolutionary algorithm.

[0111] b. Model Evaluation Metrics: Define a model adaptation function, where accuracy (such as the inverse function of RUL prediction error or fault detection accuracy) and metrics such as Belief-Reliability jointly constitute the objective. Belief-Reliability is a reliability evaluation metric that comprehensively considers uncertainty; in this context, it can be understood as an assessment of the confidence / reliability of the model in its predictions, potentially achieved through the credibility score of the prediction interval. Therefore, the adaptation function can be set with multiple objectives (maximizing accuracy and confidence) to balance model performance and robustness.

[0112] c. Parallel Simulation Evaluation: Each model is evaluated using the data / environment generated by the simulation platform. For example, for each candidate PHM model, a batch of Monte Carlo scenarios (or samples from a previously generated dataset) are run to allow the model to predict RUL or diagnose faults. The model then calculates its prediction accuracy and belief-reliability metrics. Because the population size and number of scenarios can be large, this step relies heavily on parallel computing. Ray is used to distribute the evaluation tasks for parallel execution, significantly reducing evaluation time.

[0113] d. Selection and Reproduction: Based on the evaluation results, evolutionary algorithms are applied to the model population (e.g., selecting models with high fitness for the next generation and eliminating inferior models; performing crossover and mutation on the selected models to generate new models). A multi-objective evolutionary algorithm (such as NSGA-II) can be used to prioritize both accuracy and reliability, or a simple GA can be used to weight both into a single score.

[0114] e. Automatic iteration: The above evaluation and reproduction steps are repeated to continuously generate new models. The best models are selected through simulation testing and feedback until the termination condition (algebraic upper limit or performance convergence) is met. The best performing PHM model is finally output.

[0115] Process integration method: To achieve the above closed loop, it is necessary to solve the interface integration between the simulation platform and the evolutionary algorithm framework:

[0116] a. Training control script: Use Python to write a master control script that encapsulates simulation platform calls (generating data or running a full simulation) into easy-to-use functions / services. For example, provide a run_simulation(config) ->results interface to return evaluation results given model and scenario parameters.

[0117] b. Evolutionary Algorithm Library: Existing libraries can be used (such as DEAP, Nevergrad, or Ray's built-in Tune module, which supports evolution). These libraries facilitate population management and iteration in Python. The master control script is responsible for calling the library to generate a set of model parameters in each generation, then calling the simulation evaluation function for each parameter set to obtain fitness, and then feeding the fitness back to the library for selection.

[0118] c. Distributed Architecture: For large-scale evolution, Ray's actor and task models can be leveraged to distribute simulation tasks for different model-scenario combinations to multiple workers for parallel processing. Ray can also manage GPU resources to accelerate PHM model inference training. If simulations are time-consuming, Ray Tune's trial parallelism mechanism can easily schedule hundreds of concurrent simulations to explore the model space.

[0119] Result analysis and adaptive iteration: Adaptive mechanisms can be added during the integration process:

[0120] a. If you monitor the diversity of model performance during the evolution process, if the population converges prematurely, you can introduce new random individuals to increase diversity.

[0121] b. Adjust the distribution of simulation scenarios based on simulation feedback (so-called "targeted evolution"). For example, if the model performs poorly in a certain type of scenario, more data on that type of scenario will be generated for subsequent focus, thereby guiding the algorithm to improve weak links.

[0122] c. Record the performance and characteristics of each generation of models to analyze the evolutionary path and reliability improvement process, thereby providing insights into PHM model design (i.e., achieving "model discovery" as mentioned in the question, discovering excellent model structures and configurations in simulation experiments).

[0123] Through this evolutionary training, the digital twin simulation platform not only provides data but also participates in the closed-loop model optimization, supporting the automated design and optimization of PHM models. Introducing the Belief Reliability metric in this process helps the evolutionary algorithm favor models that are both highly accurate and robust, avoiding models that overfit specific data but are not robust to uncertainty.

[0124] Other parts of the present invention that are not described in detail may refer to the existing technology or are well-known technologies to those skilled in the art. This embodiment does not limit this and will not be described in detail here.

[0125] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A method for generating PHM model training data based on digital twins, characterized in that: The method comprises: Build a medium-fidelity digital twin model that comprehensively considers the dynamic characteristics and coupling relationships of each component of the aviation system to reproduce the global dynamic behavior; The medium-fidelity digital twin model is calibrated based on expert knowledge and historical data to improve accuracy; Model simulation is achieved through Monte Carlo batch simulation, and data are labeled according to simulation conditions to generate a large amount of PHM model training data.

2. The method for generating PHM model training data based on digital twin according to claim 1, characterized in that: The PHM model training data is normal data and fault data with state labels, and the fault data includes single fault data and compound fault data.

3. The method for generating PHM model training data based on digital twin according to claim 1, characterized in that: The Monte Carlo batch simulation performs multiple simulation tests by randomly changing the initial state, operating conditions, and failure mode parameters of the aviation system, wherein the failure mode parameters include failure type, occurrence time, and severity; and injects transient / progressive failures and combined failures.

4. The method for generating PHM model training data based on digital twin according to claim 1, characterized in that: The composite fault is to inject two or more different types of faults simultaneously in one simulation test.

5. The method for generating PHM model training data based on digital twin according to claim 1, characterized in that: The medium-fidelity digital twin model includes an engine propulsion sub-model, a flight control sub-model, an avionics and sensor sub-model, and an airframe structure and dynamics sub-model.

6. The method for generating PHM model training data based on digital twin according to claim 1, characterized in that: The medium-fidelity digital twin model preserves subsystem interactions critical to PHM; including: Flight-engine coupling: the effects of aircraft attitude and speed on engine intake and loads, and the interaction of engine thrust reactions to flight conditions; Coupling of power generation and consumption: the impact of engine speed on generator output, and the impact of electrical faults on flight control and instrumentation; The impact of structural degradation on performance: increased drag of the aircraft and the impact of structural deformation on handling stability.

7. The method for generating PHM model training data based on digital twin according to claim 6, characterized in that: The influence of structural degradation on performance is reflected in medium fidelity by using parameter perturbation. This includes reducing the slope of the lift curve by introducing a health factor and characterizing the impact of structural aging on flight performance by increasing the drag coefficient.

8. The method for generating PHM model training data based on digital twins according to claim 1, characterized in that: The dynamic characteristics of each component include: Flight dynamics: uses an open source 6-degree-of-freedom nonlinear engine to simulate attitude and trajectory; Engine and propulsion systems: Modeling thermodynamics, rotor dynamics, and thrust / fuel consumption using Modelica; Power supply and other systems: Use Modelica to build models of the gas source and hydraulic subsystems to capture key physical evolutions.

9. The method for generating PHM model training data based on digital twin according to claim 1, characterized in that: The medium-fidelity digital twin model simplifies non-essential details, including: Detailed blade stress distribution inside the engine; The electrical system is simulated using an equivalent gas circuit model; The aerodynamics of the aircraft configuration uses lookup table data.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the content of the method for generating PHM model training data based on digital twins as described in any one of claims 1 to 9.

Images