Method and system for optimizing quality characteristics of an aircraft product based on uq analysis

Abstract

The application relates to the technical field of digital design and manufacturing, and discloses an aviation product quality characteristic optimization method and system based on UQ analysis. The method comprises the following steps: S1, constructing an MBSE system architecture model, and establishing a cross-level constraint relationship of aviation product quality characteristics; S2, constructing a quality characteristic-oriented digital twin model; S3, performing quality characteristic probability propagation analysis; S4, acquiring state data to dynamically calibrate the cognitive bias of the digital twin model; S5, performing global sensitivity analysis and quality risk assessment to obtain key quality influence parameters; and S6, optimizing the key quality influence parameters and obtaining an optimized aviation product quality characteristic evaluation result. The device comprises an MBSE probability architecture modeling module, a digital twin base module, an uncertainty quantification analysis module and a special solver interface module. The application realizes cross-life cycle closed-loop optimization of aviation product quality characteristics.

Description

Technical Field

[0001] This invention relates to the field of digital design and manufacturing technology, and specifically to a method and system for optimizing the quality characteristics of aerospace products based on UQ analysis. Background Technology

[0002] As the complexity of modern aviation equipment continues to increase, its design, manufacturing, and operation face significant impacts from multiple sources of uncertainty—including fluctuations in design parameter distribution, geometric deviations in manufacturing and assembly, random disturbances in test loading, dynamic changes in the service environment, and cognitive biases in simulation models. These uncertainties are coupled with each other and propagate across levels, making it difficult for traditional engineering methods based on deterministic assumptions to accurately assess the true quality characteristics of equipment (such as performance margins, safety boundaries, and mission adaptability), thereby affecting design iteration efficiency, manufacturing cost control, and service safety.

[0003] To address these issues, the industry has developed technologies such as Model-Based Systems Engineering (MBSE), Digital Twins (DT), and Uncertainty Quantification (UQ). MBSE uses system architecture models to achieve a structured expression of requirements and designs; digital twins support state perception and prediction by constructing virtual mirrors of physical assets; and UQ uses probabilistic methods to quantify the impact of input fluctuations and model cognitive biases on output responses. However, these three technologies are currently fragmented in practice: MBSE tools remain at the architectural modeling stage of the design phase, lacking dynamic data linkage with digital twins; digital twin platforms are mostly deterministic mirrors without embedded UQ computing engines; and UQ analysis exists as offline scripts or standalone tools, disconnected from the systems engineering process.

[0004] While some technologies have attempted to combine these two approaches, for example, related literature explores the practical application of digital twins through the MBSE method; other studies have proposed an integrated framework combining deep learning and MBSE to construct an adaptive manufacturing digital twin system. Chinese patent CN121920190A discloses a method for full lifecycle risk assessment of aerospace systems based on a triple V-engineering model, which uses outer, middle, and inner V-models for demand decomposition, uncertainty quantification, and knowledge evolution, respectively. However, this approach uses the V-model as a hierarchical framework, limiting the integration depth of UQ and digital twins, and failing to achieve organic integration of the three within a unified platform. Other solutions have achieved partial integration of UQ and digital twins. For instance, Chinese patent CN115238394B discloses a digital twin modeling method for the mixed reliability of multi-source uncertainties in composite material structures, employing a multi-scale high-fidelity finite element model and a polynomial chaotic expansion proxy model to achieve mixed uncertainty quantification for specific structural components. However, this approach is limited to a single structural component object, does not involve a system-level MBSE architecture model, and cannot support uncertainty propagation quantification between multiple levels. Chinese patent application CN121543214A discloses an adaptive generative digital-real fusion test method for aero-engines, focusing on the test verification stage and not covering the integration of design, manufacturing, and service stages. Chinese patent application CN121680325A discloses a flexible control method for a digital twin manufacturing system addressing uncertainties in the production process, focusing only on uncertainty compensation during the manufacturing stage.

[0005] In summary, existing technologies share the following shortcomings: First, MBSE, digital twins, and UQ are isolated from each other, or only combined in pairs with limited depth, lacking a unified platform for deep integration of the three; Second, most solutions focus on a single lifecycle stage, failing to achieve closed-loop data flow and cross-stage adaptive feedback throughout the entire lifecycle; Third, existing platforms are mostly "deterministic mirrors," outputting a single deterministic value, lacking an embedded probabilistic computing engine to quantify the uncertainty propagation path and the confidence interval of prediction results, making it difficult to support probability-based quality characteristic assessment and decision-making. Summary of the Invention

[0006] To overcome the problems in existing technologies where model-based systems engineering, digital twins, and uncertainty quantification analysis are independent, making it difficult to achieve a unified probabilistic expression and dynamic update of the quality characteristics of complex aerospace products in the design, manufacturing, and service stages, as well as the low propagation efficiency in high-dimensional random spaces, the difficulty in continuously calibrating model cognitive biases, and the difficulty in dynamically assessing quality risks across the product lifecycle, this invention provides an aerospace product quality characteristic optimization method and system based on UQ analysis to achieve cross-level probabilistic modeling, dynamic propagation, and closed-loop optimization of the quality characteristics of complex aerospace products.

[0007] Specifically, this invention provides a method for optimizing the quality characteristics of aviation products based on UQ analysis, which includes: S1: Construct an MBSE system architecture model that integrates probabilistic attributes and establish cross-level constraint relationships for the quality characteristics of aviation products; First, establish the MBSE system architecture model for aviation products. The MBSE system architecture model typically includes a scenario model, a functional model, a logical architecture model, a physical structure model, and a quality characteristic model. Then, the parameters in the MBSE system architecture model are extended with probabilistic attributes. Specifically, based on the source of uncertainty, the parameters are divided into four categories: accidental uncertainty parameters, model cognitive uncertainty parameters, deterministic design variables, and task scenario parameters. Accidental uncertainty parameters and model cognitive uncertainty parameters are collectively referred to as uncertainty parameters. Finally, a cross-level probabilistic constraint relationship is established between system-level quality characteristics, subsystem functional indicators and underlying physical parameters to form a quality characteristic probability transmission chain, and the cross-level probability transmission relationship is obtained based on the quality characteristic probability transmission chain. S2: Based on the MBSE system architecture model, construct a digital twin model oriented towards quality characteristics to realize dynamic mapping between the physical state of the product and the state of the virtual model; Based on the functional structure relationships and quality characteristic constraints defined in the MBSE system architecture model, a digital twin model corresponding to the physical product is constructed; the digital twin model is used to realize the real-time mapping and dynamic updating of quality characteristics by manufacturing deviations, process execution status and service environment disturbances. S3: Based on the uncertainty parameters in S1, construct a proxy model based on the design simulation data, and perform a quality characteristic probability propagation analysis based on the design phase information; Model the uncertainty parameters in S1; Based on the parameter dimension and variable coupling characteristics in the cross-level probability transit relationship, an adaptive experimental design method is used to construct a sample set, and a cross-fidelity surrogate model is constructed by combining a high-fidelity physical field analysis model and a low-fidelity fast prediction model; uncertainty propagation analysis is performed using the cross-fidelity surrogate model to calculate the probability distribution characteristics and reliability index of the top-level quality characteristics. S4: Acquire status data during the manufacturing, testing, and service phases, synchronize the virtual model with the physical state through a digital twin model, and dynamically calibrate the cognitive bias of the digital twin model using a Bayesian update mechanism. Acquire status data, including geometric measurement deviations, material performance parameters, and assembly process status data during the manufacturing stage, and map the status data into the digital twin model; acquire external loads, working environment, and system response timing data during the testing and service stages, and update the status of the digital twin model synchronously. Using state data as evidence of physical state observation, the model cognitive uncertainty parameters are dynamically corrected using the Bayesian update method; based on the updated posterior distribution of the model cognitive uncertainty parameters, the mass characteristic probability propagation analysis in S3 is re-executed, and then S5 is executed. S5: Perform global sensitivity analysis to obtain key quality impact parameters; S6: Optimize key quality impact parameters and obtain the optimized evaluation results of aviation product quality characteristics.

[0008] Preferably, in S1, the cross-level probability transmission relationship expression obtained from the quality characteristic probability transmission chain is as follows: ; in, Indicates the system-level quality characteristic response. The parameter representing random uncertainty with a probability distribution. This represents the model's cognitive uncertainty parameter. Design variables representing deterministic values, Indicates task scenario parameters. This represents the cross-level logic transfer function defined in the MBSE architecture.

[0009] Preferably, in S3, based on the parameter dimension and variable coupling characteristics in the cross-level probability transitive relationship, an adaptive experimental design method is used to construct a sample set, and a cross-fidelity surrogate model is constructed by combining a high-fidelity physics field analysis model and a low-fidelity fast prediction model. The mathematical expression of the cross-fidelity surrogate model is as follows: ; in, For high-fidelity response prediction values, For low-fidelity response predictions, This is the scale scaling factor. A bias Gaussian process is used to capture the nonlinear differences between high-fidelity and low-fidelity models; The input vector represents the set of key design parameters, process parameters, and environmental uncertainty parameters that affect the system response characteristics; Uncertainty propagation analysis is performed using a cross-fidelity surrogate model, and the probability distribution characteristics and reliability indices of top-level quality properties are calculated using polynomial chaotic expansion (PCE) or Monte Carlo (MCS) methods.

[0010] Preferably, in S4, state data is used as evidence of physical state observation, and the model's cognitive uncertainty parameters are dynamically corrected using a Bayesian update method. The specific formula for the Bayesian update mechanism is as follows:

[0011] in, To provide physical state observation evidence for the synchronization of digital twin models. For the prior distribution of the model's cognitive uncertainty parameters, Given the model, cognitive uncertainty parameters The likelihood function of the observed data at time, The posterior distribution of the cognitive uncertainty parameters of the dynamically calibrated model; Based on the updated model's understanding of the posterior distribution of uncertainty parameters, the S3 quality characteristic probability propagation analysis is performed again.

[0012] Preferably, the aviation product is a hydraulic actuator in the flight control system.

[0013] Preferably, the cross-level probability transitivity relationship in S1 is as follows: ; in, This represents the system-level quality characteristic response, in mission reliability analysis. Flight control system mission reliability In the analysis of instruction tracing capabilities, Take the system-level displacement tracking error or actuation capability margin. This represents a cross-level transfer function formed by connecting contact mechanics models, friction and wear models, leakage models, hydraulic actuator dynamics models, or mission reliability models according to the MBSE architecture; Manufacturing and assembly fluctuation parameters include piston rod diameter deviation, cylinder inner diameter deviation, seal ring size deviation, seal groove size deviation, surface roughness deviation, coaxiality deviation, and assembly compression deviation. This refers to the batch dispersion of material property fluctuation parameters, including the elastic modulus, hardness, Poisson's ratio, and friction coefficient parameters of sealing materials. This includes environmental and mission disturbance parameters, such as fluctuations in hydraulic pressure, ambient temperature, external loads, actuation frequency, and mission load spectrum parameters. and Together they constitute the uncertainty parameters : The contact model cognitive parameters include contact stiffness correction coefficients, boundary constraint correction coefficients, and material constitutive model deviations; The wear model uses cognitive parameters, including wear coefficient, temperature correction coefficient, and degradation rate correction coefficient. The parameters for understanding leakage and hydraulic dynamics models include effective clearance correction factor, leakage model deviation, friction resistance correction factor, and valve control dynamic parameter deviation; , and Together they form the model of cognitive uncertainty parameters ; These are deterministic design variables, including the nominal inner diameter of the cylinder, the nominal diameter of the piston rod, the nominal cross-sectional dimensions of the sealing ring, the nominal sealing clearance, and the design values ​​for surface roughness. These are the parameters for the task scenario, including the task phase, duration, command amplitude, and the number of specified actions.

[0014] Preferably, the digital twin model in S2 includes a structural geometric twin, a physical behavior twin, and a state data twin, used to receive real-time state data collected during the manufacturing, testing, and service phases, and to realize the dynamic mapping of the physical product state to the virtual model state. Among them, the structural geometric twin simultaneously preserves the nominal design geometry and the measured geometry of a single part, including the cylinder inner diameter, piston rod diameter, sealing ring cross-sectional dimensions, sealing groove dimensions, sealing gap, surface roughness, coaxiality, and assembly compression, so that the twin model reflects the real geometric state of the single product. The physical behavior twin consists of a contact mechanics model, a friction and wear model, a leakage model, a hydraulic actuator dynamics model, and a task reliability model. Each model reads the corresponding parameters in the cross-level probability transfer relationship and calls an external solver through the analysis model integrated within the MBSE system architecture model or through a dedicated solver interface module. After analysis and solving, the output includes contact pressure, contact deformation, wear depth, leakage amount, output force, response time, displacement tracking error, and task reliability. State data twins are used to receive data from the manufacturing, testing, and service phases, and store them according to the source of the parameters to avoid confusion between measured random fluctuations and model-copied parameters; The digital twin model enables real-time mapping and dynamic updating of quality characteristics based on manufacturing deviations, process execution status, and service environment disturbances.

[0015] Preferably, in S3, for random uncertainty parameters, a probability distribution function is used for modeling. For model cognitive uncertainty parameters, interval probability, prior probability distribution, or evidence structure are used to describe them based on existing validation data. Within the input space composed of the above parameters, the Latin hypercube method is used to generate initial samples, and high-fidelity and low-fidelity models are established according to different responses. The high-fidelity models include a nonlinear contact finite element model of the sealing ring, a thermo-fluid-structure interaction leakage model, and a hydraulic actuator system model that includes valve control and pipeline dynamic characteristics. The low-fidelity models include a simplified sealing compression model, a rapid wear model, an equivalent gap leakage analytical model, and a reduced-order actuator dynamic model. A cross-fidelity surrogate model is established for each type of quality characteristic. After training the cross-fidelity surrogate model, high-fidelity samples are adaptively increased based on the cross-validation error and the Gaussian process prediction variance until each response meets the preset accuracy. The cross-fidelity surrogate model that has been validated is called through the Monte Carlo method to complete the probability propagation.

[0016] Preferably, in S4, for hydraulic actuators in the flight control system, geometric measurement deviations, material performance parameters, and assembly process status data from the manufacturing stage are acquired, as well as pressure, temperature, load, displacement, output force, response time, leakage, wear, and cumulative actuation count data from the testing and service stages. After unit unification, outlier processing, and time alignment, the actual manufacturing measurements are used to update the manufacturing and assembly fluctuation parameters in the digital twin model. and material property fluctuation parameters The operating time series is used to update the environmental and task perturbation parameters in the digital twin model. and task scenario parameters The response data serves as physical state observation evidence for calibrating the model's cognitive uncertainty parameters; Based on the sources of the model's cognitive uncertainty parameters, the physical state observation evidence is divided into three groups; Physical observational evidence representing the physical state of contact behavior, including contact pressure, contact deformation, or seal compression; Observational evidence of the physical state of wear and degradation, including wear depth, mass loss, and degradation trajectory as a function of the number of cycles; Evidence of leakage and actuation performance based on physical observations, including leakage amount, output force, response time, and displacement feedback. , and Perform a joint Bayesian update: ; In the formula, To access the cognitive parameters of the model, For wear model cognitive parameters, For understanding parameters of leakage and hydraulic dynamics models, This is a combined vector of all cognitive uncertainty parameters for the model. For the combination of all physical state observation evidence, , and These are the likelihood functions for the cognitive uncertainty parameters of the contact model, the wear model, and the leakage and hydraulic dynamics model, respectively. It is a joint prior distribution; It is a joint posterior distribution.

[0017] Furthermore, the present invention also provides a system for optimizing the quality characteristics of aviation products based on UQ analysis, including: an MBSE probabilistic architecture modeling module, a digital twin base module, an uncertainty quantification analysis module, and a dedicated solver interface module; The MBSE probabilistic architecture modeling module is used to construct a probabilistic system architecture model for evaluating the quality characteristics of aerospace products. It includes a hierarchical structure decomposition and model building submodule, a quality characteristic correlation construction submodule, a key quality characteristic parameter definition and description submodule, and an automatic parameter information mapping submodule. The MBSE probabilistic architecture modeling module realizes the cross-level probabilistic expression of the quality characteristics of aerospace products by modeling the quality constraint relationships between the system level, subsystem level, component level, and physical parameter level, and provides a unified parameter relationship foundation for subsequent uncertainty propagation analysis and dedicated solver calls. The digital twin foundation module provides data management, computing support, visualization, and basic functional support, including a database submodule, a general computing submodule, a visualization submodule, and a basic functional submodule. The digital twin foundation module stores and manages product design data, simulation data, manufacturing data, test data, and service status data, and supports dynamic mapping and visualization between physical product status, virtual model status, and quality characteristic evaluation results. The uncertainty quantification analysis module is used to perform uncertainty modeling, propagation analysis, and post-processing of the quality characteristics of complex aerospace products. It includes a simulation and external interface call submodule, a random input modeling and sampling submodule, a surrogate model construction submodule, and a post-processing submodule. The uncertainty quantification analysis module is used to identify uncertainty factors such as manufacturing fluctuations, material dispersion, environmental disturbances, and model cognitive biases, construct random input models, and generate sampling samples. It obtains response results by calling simulation models or external solvers, constructs surrogate models, and performs quality characteristic probability propagation, reliability index calculation, sensitivity analysis, and quality risk assessment. The dedicated solver interface module is used to realize data interaction and task scheduling with different types of engineering solvers and computational tools, including cross-level and cross-stage system simulation solvers, finite element solvers, multiphysics coupling solvers, numerical computation and mathematical analysis solvers, optimization design solvers, and parallel computing support modules. This module is used to call the corresponding structural analysis, system simulation, multiphysics coupling, mathematical computation, and optimization design tools according to the analysis requirements of the quality characteristic assessment task, and improves the computational efficiency of high-dimensional uncertainty propagation and multi-condition quality assessment through the parallel computing support module. The hierarchical constraints, key quality characteristic parameters, and parameter mapping relationships output by the MBSE probabilistic architecture modeling module are passed to the uncertainty quantification analysis module. The uncertainty quantification analysis module calls the corresponding engineering solver through the dedicated solver interface module to complete sample response calculation, surrogate model training, and quality risk assessment. The digital twin base module provides data storage, state mapping, calculation management, and result visualization support for the above modules, thereby forming an integrated system for probabilistic modeling, dynamic propagation, simulation solving, and evaluation feedback for the quality characteristics of complex aerospace products.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a collaborative mechanism of "MBSE probabilistic architecture - digital twin state mapping - uncertainty quantification dynamic propagation" to achieve unified probabilistic assessment and dynamic updating of the quality characteristics of complex aerospace products throughout the design, manufacturing and service processes, thereby improving the ability to predict quality risks and the credibility of assessments, and realizing closed-loop optimization of the quality characteristics of complex aerospace products across their life cycle. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the steps of the aviation product quality characteristic optimization method based on UQ analysis of the present invention. Figure 2 This is a schematic diagram of the hierarchy of hydraulic actuators in the flight control system of this invention; Figure 3 This is a schematic diagram of the system modules of the aviation product quality characteristic optimization method based on UQ analysis of the present invention. Detailed Implementation

[0020] The embodiments of the present invention will be described below with reference to the accompanying drawings. Addressing the problems of fragmented data across stages, difficulty in uniformly describing the coupling relationships of quality characteristics across levels, difficulty in quantifying and propagating uncertainties, and high computational costs of high-fidelity simulation in complex aerospace products throughout the design, manufacturing, testing, and service processes, the present invention provides a method for optimizing the quality characteristics of aerospace products based on UQ analysis.

[0021] This embodiment uses the hydraulic actuator in the flight control system as a typical example to illustrate the optimization method for aviation product quality characteristics based on UQ analysis. Specifically, as follows... Figure 1 As shown, the steps include the following.

[0022] S1: Construct an MBSE system architecture model that integrates probabilistic attributes and establish cross-level constraint relationships for the quality characteristics of aviation products.

[0023] First, establish an MBSE system architecture model for aviation products. The MBSE system architecture model typically includes a scenario model, a functional model, a logical architecture model, a physical structure model, and a quality characteristic model.

[0024] Then, the parameters in the MBSE system architecture model are extended with probabilistic attributes. Specifically, based on the source of uncertainty, the parameters are divided into four categories: accidental uncertainty parameters, model cognitive uncertainty parameters, deterministic design variables, and task scenario parameters. All four categories of parameters are multi-source parameters. Among them, accidental uncertainty parameters and model cognitive uncertainty parameters are collectively referred to as uncertainty parameters, while deterministic design variables and task scenario parameters are deterministic parameters.

[0025] Finally, a cross-level probabilistic constraint relationship is established between system-level quality characteristics, subsystem functional indicators, and underlying physical parameters to form a quality characteristic probability transmission chain. Based on the quality characteristic probability transmission chain, the expression for the cross-level probability transmission relationship is as follows: ; in, Indicates the system-level quality characteristic response. The parameter representing random uncertainty with a probability distribution. This represents the model's cognitive uncertainty parameter. Represents deterministic design variables. Indicates task scenario parameters. This represents the cross-level logic transfer function defined in the MBSE architecture.

[0026] Based on the MBSE system architecture model, this paper analyzes the input parameters, boundary conditions, and response output of the underlying physical field analysis model in the scenario model, establishes a mapping relationship between the system architecture parameters and the physical field model parameters, and realizes the correlation between system-level mass characteristics and underlying physical responses. After establishing the parameter mapping relationship, simulation analysis can be performed by calling an external solver through the internal integrated analysis module or through a dedicated solver interface module to obtain the corresponding output.

[0027] This embodiment takes the hydraulic actuator in the flight control system as a specific object and establishes an MBSE system architecture model for the hydraulic actuator, including a scenario model, a functional model, a logical architecture model, a physical structure model, and a quality performance model. The scenario model describes the working conditions of the flight control system during mission phases such as takeoff, cruise, maneuvering, and landing. The functional model describes the functional links such as flight control command generation, signal transmission, actuator execution, and feedback control. The logical architecture model describes the logical components, information interaction relationships, control relationships, and functional allocation relationships required for the flight control system to complete various functions, without limiting specific product models or physical component forms. The physical structure model describes the actual physical components, assembly relationships, interface relationships, material properties, geometric parameters, and working mechanisms for implementing logical functions. The quality performance model describes quality characteristic indicators such as actuator displacement accuracy, response time, output force, sealing performance, wear life, and mission reliability.

[0028] The hierarchy of hydraulic actuators in a flight control system, such as Figure 2 As shown, the system is decomposed into a hierarchy of "flight control system - hydraulic actuator - sealing assembly - piston rod and seal ring", so that the system-level mission reliability requirements can be linked layer by layer to the actuator output force, response time, displacement tracking error, seal leakage, seal wear and remaining life.

[0029] When extending the probabilistic attributes of parameters at each layer, uncertainties from different sources are no longer uniformly written into a single parameter without distinguishing their sources. Instead, parameter sets are set separately according to the generation mechanism.

[0030] Among them, the random uncertainty parameter Including: Manufacturing and assembly fluctuation parameters denoted as This includes piston rod diameter deviation, cylinder inner diameter deviation, seal ring size deviation, seal groove size deviation, surface roughness deviation, coaxiality deviation, and assembly compression deviation; material property fluctuation parameters are denoted as... This includes batch dispersion of parameters such as the elastic modulus, hardness, Poisson's ratio, and coefficient of friction of the sealing material; environmental and mission disturbance parameters are denoted as... This includes fluctuations in parameters such as hydraulic pressure, ambient temperature, external load, actuation frequency, and mission load spectrum. These parameters exhibit repeatable randomness, and probability distributions and correlations are established based on test, experimental, or service samples.

[0031] For model cognitive uncertainty parameters Based on the corresponding physical processes, it is further divided into: contact model cognitive parameters denoted as This includes contact stiffness correction factors, boundary constraint correction factors, and material constitutive model deviations; wear model cognitive parameters are denoted as... This includes the wear coefficient, temperature correction coefficient, and degradation rate correction coefficient; leakage and hydraulic dynamics model cognitive parameters are denoted as... This includes the effective clearance correction factor, leakage model deviation, friction resistance correction factor, and valve control dynamic parameter deviation.

[0032] Deterministic design variables are denoted as This includes the nominal inner diameter of the cylinder, the nominal diameter of the piston rod, the nominal cross-sectional dimensions of the sealing ring, the nominal sealing clearance, and the design values ​​for surface roughness.

[0033] Task scenario parameters are denoted as This includes the task phase, duration, instruction amplitude, and the number of specified actions.

[0034] Accordingly, this embodiment concretizes the cross-level probability transfer relationship in the claims as follows: ; In the formula, This represents the system-level quality characteristic response, in mission reliability analysis. Flight control system mission reliability In the analysis of instruction tracing capabilities, Take the system-level displacement tracking error or actuation capability margin. This represents a cross-level transfer function formed by connecting contact mechanics models, friction and wear models, leakage models, hydraulic actuator dynamics models, or mission reliability models according to the MBSE architecture.

[0035] Each of the above parameter sets corresponds one-to-one with a specific calculation item. This includes manufacturing and assembly fluctuation parameters. Dimensional and assembly deviations, material property fluctuation parameters Material elastic modulus and friction coefficient, contact model cognitive parameters Contact model correction parameters and The nominal geometric dimensions are input into the contact mechanics model to obtain the contact pressure and contact deformation of the sealing ring. Next, the contact pressure, friction coefficient, relative sliding speed, temperature, and wear model parameters of the sealing ring are input. The wear model parameters are input into the wear model to obtain the wear depth and its evolution with the number of actuations. Then, the effective clearance, hydraulic pressure, leakage, and hydraulic dynamics model parameters after wear are input into the model. The leakage correction parameters are input into the leakage model to obtain the leakage amount. Subsequently, Frictional resistance, oil supply pressure, and valve control parameters are input into the dynamic model of the hydraulic actuator to obtain the output force. Response time and displacement tracking error Finally, based on the component parameters related to the mission reliability model, the parameters obtained from the contact mechanics model, friction and wear model, leakage model, and hydraulic actuator dynamics model, the deterministic design variables, and the mission scenario parameters, all are input into the mission reliability model to obtain the evaluation results of the aviation product quality characteristics.

[0036] To enable actual calls between models, a mapping dictionary is established between MBSE parameter nodes and finite element models, multiphysics models, and system simulation models. The mapping dictionary records the unique identifier of each parameter, its associated parameter set, the variable name on the model side, its unit, data type, input or output attribute, and its hierarchical source. For example, piston rod diameter deviation can be identified as a manufacturing and assembly fluctuation parameter. In the geometric input, the elastic modulus of the sealing material is identified as... The material input in the file will identify the wear coefficient as... The parameters to be calibrated are recorded, and the contact pressure, wear depth, leakage amount, and actuation performance results are written back to the corresponding quality characteristic nodes, thereby forming an executable and traceable probability transmission chain.

[0037] S2: Based on the MBSE system architecture model, construct a digital twin model oriented towards quality characteristics to achieve dynamic mapping between the physical state of the product and the state of the virtual model.

[0038] Based on the functional structural relationships and quality characteristic constraints defined in the MBSE system architecture model, a digital twin model corresponding to the physical product is constructed. This digital twin model includes a structural geometric twin, a physical behavioral twin, and a state data twin. It is used to receive real-time state data collected during the manufacturing, testing, and service phases, and to achieve dynamic mapping from the physical product state to the virtual model state. The digital twin model enables real-time mapping and dynamic updating of quality characteristics based on manufacturing deviations, process execution status, and service environment disturbances.

[0039] For the hierarchical architecture, parameter set, and model mapping relationship of the hydraulic actuators in the flight control system based on S1, a digital twin model of the hydraulic actuator corresponding to the actual hydraulic actuator is constructed. This digital twin model includes structural geometric twins, physical behavioral twins, and state data twins, specifically: The structural geometric twin simultaneously preserves both the nominal design geometry and the measured geometry of individual components, including cylinder bore diameter, piston rod diameter, seal ring cross-sectional dimensions, seal groove dimensions, seal clearance, surface roughness, coaxiality, and assembly compression. After manufacturing and inspection are completed, the measured values ​​for the specific product are written into... The corresponding data nodes enable the twin model to reflect the true geometric state of a single product.

[0040] The physical behavior twin consists of contact, wear, leakage, actuator dynamics, and mission reliability models in S1. Each model is read separately. , , , , , , and The corresponding parameters are obtained, and the external solver is called through the internal integrated analysis model or through the dedicated solver interface module. After analysis and solution, the contact pressure, contact deformation, wear depth, leakage, output force, response time, displacement tracking error and task reliability are output.

[0041] State data twins are used to receive data from the manufacturing, testing, and service phases, and store them according to the source of the parameters to avoid confusion between the random uncertainty parameters of measured random fluctuations and the uncertainty parameters of model cognition.

[0042] For the manufacturing stage data of hydraulic actuators in the flight control system, this includes dimensional deviations, surface condition, and assembly status; for the testing stage data, it includes hydraulic pressure, temperature, displacement, output force, response time, leakage, and wear; and for the service stage data, it includes mission profiles, actuation frequency, load spectrum, pressure and temperature timing, displacement feedback, fault records, and cumulative operating time. After the data is entered into the digital twin model, instance binding and time alignment are first performed based on the product number, measurement point number, and timestamp; subsequently, the geometric and assembly data are updated to... Update the material re-inspection data to Update the pressure, temperature, and load timing data to and Finally, call the function in S1. Recalculate the quality characteristics of each layer.

[0043] When a stable deviation occurs between the measured response and the virtual model prediction, this deviation is not directly incorporated into the random fluctuation parameters. Instead, it is used as the basis for subsequent calibration based on the physical process corresponding to the deviation. , or The observational basis. For example, contact pressure deviation is used for calibration. Wear depth deviation is used for calibration. Leakage and actuation performance deviations are used for calibration. This enables a clear distinction and dynamic mapping between physical states, random uncertainty parameters, and model cognitive uncertainty parameters.

[0044] S3: Based on the uncertainty parameters in S1, construct a proxy model based on the design simulation data, and perform a quality characteristic probability propagation analysis based on the design phase information.

[0045] Different methods are used to model the uncertainty parameters in S1: probability distribution functions are used to model the random uncertainty parameters, while interval probability, prior probability distribution, or evidence structure are used to describe the model's cognitive uncertainty parameters.

[0046] Based on the parameter dimension and variable coupling characteristics in cross-level probability transitivity, an adaptive experimental design method is used to construct a sample set, and a cross-fidelity surrogate model is constructed by combining a high-fidelity physics analysis model and a low-fidelity fast prediction model. The mathematical expression of the cross-fidelity surrogate model is as follows: ; in, For high-fidelity response prediction values, For low-fidelity response predictions, This is the scale scaling factor. A bias Gaussian process is used to capture the nonlinear differences between high-fidelity and low-fidelity models; The input vector represents the set of key design parameters, process parameters, and environmental uncertainty parameters that affect the system's response characteristics.

[0047] Uncertainty propagation analysis is performed using a cross-fidelity surrogate model, and the probability distribution characteristics and reliability indices of the top-level quality characteristics are calculated using either polynomial chaotic expansion (PCE) or Monte Carlo (MCS) methods. If the PCE method is used, the quality characteristic response... Approximate expression: ; in, This is how to obtain a vector of orthogonal random variables from input random variables through probability transformation. For multidimensional orthogonal polynomial basis functions, Let be the expansion coefficients to be solved. This is the number of terms to be truncated during expansion.

[0048] For hydraulic actuators in the flight control system, specifically: According to S1, since , and For parameters that are subject to random uncertainty, a probability distribution function is used for modeling. For parameters with sufficient samples, a normal distribution, log-normal distribution, Weibull distribution, or uniform distribution is selected based on the goodness of fit. Dimensional deviations, material parameters, or operating condition parameters that are correlated are characterized by a joint probability distribution or correlation matrix. , and These parameters represent uncertainties in the model's cognition and are described using interval probabilities, prior probability distributions, or evidence structures based on existing validation data. Within the input space comprised of these parameters, initial samples are generated using the Latin hypercube method, and high-fidelity and low-fidelity models are established based on different responses. High-fidelity models include a nonlinear contact finite element model of the sealing ring, a thermo-fluid-structure interaction leakage model, and a hydraulic actuator system model incorporating valve control and pipeline dynamics. Low-fidelity models include a simplified seal compression model, a rapid wear model, an equivalent gap leakage analytical model, and a reduced-order actuator dynamics model.

[0049] To avoid using the same scaling parameter to describe outputs with different properties, a cross-fidelity surrogate model is established for each type of quality characteristic: ; In the formula, superscript Indicates the specific response type. Corresponding contact pressure Corresponding wear depth Corresponding leakage amount The corresponding output force at that time The corresponding response time The corresponding displacement tracking error. and The first High-fidelity and low-fidelity calculation results for class responses; This is the scaling factor corresponding to the response; This represents the deviation Gaussian process corresponding to this response. Therefore, contact pressure, wear, leakage, and dynamic performance each have independent characteristics. and They do not share the same undifferentiated unified model.

[0050] Given the input vector, by the first... It consists of a subset of parameters related to the class response. For example, the contact pressure proxy model mainly reads... Dimensions and assembly parameters in elastic parameters in and The wear proxy model further reads this information. Temperature and operating conditions in the middle and The leakage proxy model primarily reads the clearance, pressure, and... after wear. The actuation performance surrogate model reads leakage, frictional resistance, hydraulic pressure, and valve control parameters. After training, the surrogate model adaptively increases the number of high-fidelity samples based on cross-validation error and Gaussian process prediction variance until each response meets the preset accuracy.

[0051] Probability propagation is conducted using a validated surrogate model. When employing the PCE method, expansions are established for different quality characteristics:

[0052] In the formula, Indicates the first Class-specific quality characteristics. hour Leakage amount ; hour For the life of the sealing ring ; hour For output force ; hour For response time ; hour For displacement tracking error ; hour For mission reliability . For random and uncertain variables , and And the model's cognitive uncertainty parameters after probabilistic description. The vector of orthogonal random variables obtained after probability transformation; For the corresponding multidimensional orthogonal polynomial basis; The expansion coefficients are independently obtained for the k-th type of quality characteristic; Let be the number of truncation terms for the k-th type of quality characteristic.

[0053] By each The PCE model calculates the mean, variance, quantiles, confidence intervals, and probability of exceeding limits. A leakage failure event is defined as... Exceeding the permissible leakage amount, the lifetime failure event is defined as follows: The failure events of actuation performance below the required lifespan are defined as follows: Below minimum output force, response time Exceeding the maximum response time or displacement tracking error Exceeding the allowable error. If PCE is not used, the same parameter classification and failure criteria are used, and probability propagation is completed by calling the cross-fidelity surrogate model through the Monte Carlo method.

[0054] S4: Acquire status data during the manufacturing, testing, and service phases, synchronize the virtual model with the physical state through a digital twin model, and dynamically calibrate the cognitive bias of the digital twin model using a Bayesian update mechanism.

[0055] Acquire status data, including geometric measurement deviations, material performance parameters, and assembly process status data during the manufacturing stage, and map the status data into the digital twin model; acquire external loads, working environment, and system response timing data during the testing and service stages, and update the status of the digital twin model synchronously.

[0056] Using state data as evidence of physical state observation, the model's cognitive uncertainty parameters are dynamically corrected through a Bayesian update method. The specific formula for the Bayesian update mechanism is: ; in, To provide physical state observation evidence for the synchronization of digital twin models. For the prior distribution of the model's cognitive uncertainty parameters, Given the model, cognitive uncertainty parameters The likelihood function of the observed data at time, This represents the posterior distribution of the cognitive uncertainty parameters of the dynamically calibrated model.

[0057] Using the updated model to understand the uncertainty parameters, perform the quality characteristic probability propagation analysis in S3 again.

[0058] For hydraulic actuators in flight control systems, data on geometric measurement deviations, material performance parameters, and assembly process status during the manufacturing phase are acquired, along with data on pressure, temperature, load, displacement, output force, response time, leakage, wear, and cumulative actuation counts during the testing and service phases. After unit standardization, outlier handling, and time alignment, the actual manufacturing measurements are used to update the digital twin model. and Using operating condition time series to update the digital twin model and The response data serves as physical state observation evidence for calibrating the model's cognitive uncertainty parameters.

[0059] Based on the sources of the model's cognitive uncertainty parameters, the physical state observation evidence is divided into three groups. Physical observational evidence representing the physical state of contact behavior, including contact pressure, contact deformation, or seal compression; Observational evidence of the physical state of wear and degradation, including wear depth, mass loss, and degradation trajectory as a function of the number of cycles; Physical observation evidence representing leakage and operational performance includes leakage amount, output force, response time, and displacement feedback. Accordingly, for , and Perform a joint Bayesian update: ; In the formula, This is a combined vector of all cognitive uncertainty parameters for the model. This is a combined vector of all physical state observation evidence. , and These are the likelihood functions for the cognitive uncertainty parameters of the contact model, the wear model, and the leakage and hydraulic dynamics model, respectively. It is a joint prior distribution; It is a joint posterior distribution.

[0060] In practice, the contact stiffness correction factor is calibrated using contact pressure and compression residuals, etc. Parameters, such as wear coefficient and temperature correction coefficient, are calibrated using wear depth and degradation trajectory residuals. Parameters, such as leakage, output force, and response time residual calibration, include effective clearance correction factor, frictional resistance correction factor, and valve control dynamic deviation. Parameters. The likelihood function considers both sensor measurement uncertainty and model residual variance. Posterior samples are obtained through Markov chain Monte Carlo or numerical integration.

[0061] Subsequently, with Replace the prior distribution of the cognitive uncertainty parameters of each model, and , and The posterior samples are input into their corresponding contact models, wear models, and leakage and hydraulic dynamics models, respectively, and the S3 surrogate model or PCE model is called again to update. , , , , and The probability distribution is thus determined. In this way, each set of model cognitive uncertainty parameters in the formula is calibrated using data corresponding to its physical source, avoiding the direct calibration of irrelevant contact parameters with leaked data or the substitution of model cognitive bias with geometric measurement data. After re-executing the mass characteristic probability propagation analysis in S3, proceed to S5.

[0062] S5: Perform global sensitivity analysis to obtain key quality impact parameters.

[0063] The contribution of each parameter to the fluctuation of the target quality characteristics is calculated to form a global sensitivity index. The Sobol total-order sensitivity index is used. The evaluation is conducted using the following formula: ; in, For the first The total order sensitivity index of each parameter The set of all parameters involved in the propagation analysis of this quality characteristic can be derived from... , , and ; This represents the i-th parameter; Indicates except Other parameters besides those mentioned above. The total variance of the quality characteristic response, To exclude the first The expected value of the response variance caused by all parameters other than the one parameter.

[0064] Key quality-affecting parameters are identified based on the overall order sensitivity index.

[0065] For hydraulic actuators in the flight control system, based on the probability propagation results obtained in S3 and the posterior distribution of model cognitive uncertainty parameters obtained in S4, quality risk assessments are conducted on leakage, seal life, output force, response time, displacement tracking error, and mission reliability. Failure probability, reliability, confidence interval, and risk level are calculated according to the allowable boundaries of each quality characteristic.

[0066] To differentiate the influencing factors of different target quality characteristics, for each Calculate the Sobol total order sensitivity index separately: ; In the formula, Indicates input parameters For the Class quality characteristics The overall order sensitivity index; All uncertainty inputs involved in the propagation analysis of this quality characteristic can come from... , , , , , , or ; Indicates except Other inputs besides superscript. To maintain consistency with S3, the same parameter can produce different sensitivity results for different quality characteristics.

[0067] when ,Right now When analyzing leakage, the focus is on piston rod diameter deviation, cylinder inner diameter deviation, seal ring size deviation, wear depth, hydraulic pressure, and... The contribution of the effective gap correction coefficient; when ,Right now When considering the lifespan, the focus is on analyzing the friction coefficient, temperature, load spectrum, and... The contribution of the wear coefficient and degradation rate correction coefficient; At that time, the focus of analysis should be on pressure, leakage, frictional resistance, and The contribution of dynamic parameters of valve control; when At the same time, a comprehensive analysis is conducted on the contribution of task profile, actuation frequency, and quality characteristics of each subsystem.

[0068] The key quality influence parameters of the response are obtained based on the overall order sensitivity index.

[0069] S6: Optimize key quality impact parameters and obtain the optimized evaluation results of aviation product quality characteristics.

[0070] After optimizing the key quality-influencing parameters, S1 is executed again to obtain the optimized evaluation results of the aviation product quality characteristics.

[0071] Preferably, for the optimization of key quality impact parameters, optimization suggestions can be automatically provided based on the key quality impact parameters identified by S5 and the preset parameter optimization rules, and then the key quality impact parameters can be optimized according to the optimization suggestions.

[0072] Preferably, after executing S1 again, executing S2-S5 can further optimize the key quality-affecting parameters, thus achieving a closed-loop update.

[0073] This invention enables the identification of key quality-affecting parameters based on sensitivity indicators, and generates closed-loop feedback decisions for system architecture optimization, key tolerance adjustment, manufacturing compensation, and service maintenance.

[0074] For hydraulic actuators in flight control systems, sensitivity results directly correspond to feedback measures. For example, if the leakage is determined based on the sensitivity results, it is mainly affected by... If the leakage is affected by dimensional and assembly deviations, then during the manufacturing stage, measures such as tolerance tightening, parts grouping and matching, assembly compression adjustment, or machining compensation should be taken; if the leakage is mainly affected by... If the impact is significant, prioritize conducting leak tests and calibrating the model, rather than blindly adjusting product dimensions. If lifespan is primarily affected by... If the material dispersion affects the process, adjust the batch control and surface treatment; if it is mainly affected by... If the impact is felt, supplementary wear tests should be conducted and the degradation model updated. If the actuation performance is mainly affected by hydraulic pressure and valve control parameters, the pressure supply range and control parameters should be adjusted; if mission reliability is mainly affected by mission profile and remaining life, mission limitations, maintenance timing, and spare parts replacement strategies should be generated.

[0075] During the design phase, key sensitivity parameters are fed back to The nominal diameter of the piston rod, the size of the sealing ring, the sealing gap, the material, and the surface roughness were optimized for reliability and robustness; during the manufacturing stage, individual parts were measured. The risk results are used for tolerance allocation, processing compensation, grouping and matching, and re-inspection; during the service phase, the updated... , and The posterior distribution is used for remaining useful life prediction, inspection cycle adjustment, and maintenance decisions. After feedback is completed, the updated distribution will be... , , , , , , and By re-entering the transfer function of S1, the optimized quality characteristic evaluation result is obtained. If S2 to S5 are called in sequence, a closed-loop update can be achieved.

[0076] Preferably, the system output may include the probability distribution and confidence interval of key quality characteristics such as actuator actuation performance, sealing performance, remaining life of sealing rings and mission reliability, as well as the sensitivity ranking of each uncertainty parameter and the identification results of key risk parameters, and parameter optimization suggestions for improving key quality characteristics.

[0077] Furthermore, the present invention also provides a system for optimizing the quality characteristics of aerospace products based on UQ analysis, such as... Figure 3 As shown, the system includes MBSE probabilistic architecture modeling module 1, digital twin base module 2, uncertainty quantification analysis module 3, and dedicated solver interface module 4.

[0078] MBSE Probabilistic Architecture Modeling Module 1 is used to construct a probabilistic system architecture model for evaluating the quality characteristics of aerospace products. It includes sub-modules for hierarchical structure decomposition and model building, quality characteristic relationship construction, key quality characteristic parameter definition and description, and automatic parameter information mapping. By modeling the quality constraint relationships between the system level, subsystem level, component level, and physical parameter level, the MBSE probabilistic architecture modeling module achieves cross-level probabilistic expression of aerospace product quality characteristics and provides a unified parameter relationship foundation for subsequent uncertainty propagation analysis and dedicated solver calls.

[0079] The Digital Twin Base Module 2 provides data management, computational support, visualization, and basic functional support, including a database submodule, a general computing submodule, a visualization submodule, and a basic functional submodule. The Digital Twin Base Module stores and manages product design data, simulation data, manufacturing data, test data, and service status data, and supports the dynamic mapping and visualization between the physical product status, the virtual model status, and the quality characteristic evaluation results.

[0080] The Uncertainty Quantification Analysis Module 3 is used to perform uncertainty modeling, propagation analysis, and post-processing of complex aerospace product quality characteristics. It includes a simulation and external interface call submodule, a random input modeling and sampling submodule, a surrogate model construction submodule, and a post-processing submodule. The Uncertainty Quantification Analysis Module identifies uncertainties such as manufacturing fluctuations, material dispersion, environmental disturbances, and model cognitive biases; constructs a random input model and generates sampled data; obtains response results by calling simulation models or external solvers; constructs surrogate models; and performs quality characteristic probability propagation, reliability index calculation, sensitivity analysis, and quality risk assessment.

[0081] The dedicated solver interface module 4 is used to realize data interaction and task scheduling with different types of engineering solvers and computational tools, including cross-level and cross-stage system simulation solvers, finite element solvers, multiphysics coupling solvers, numerical computation and mathematical analysis solvers, optimization design solvers, and parallel computing support modules. This module is used to call upon corresponding structural analysis, system simulation, multiphysics coupling, mathematical computation, and optimization design tools according to the analytical requirements of the quality characteristic assessment task, and improves the computational efficiency of high-dimensional uncertainty propagation and multi-condition quality assessment through the parallel computing support module.

[0082] The hierarchical constraints, key quality characteristic parameters, and parameter mapping relationships output by the MBSE probabilistic architecture modeling module are passed to the uncertainty quantification analysis module. The uncertainty quantification analysis module calls the corresponding engineering solver through the dedicated solver interface module to complete sample response calculation, surrogate model training, and quality risk assessment. The digital twin base module provides data storage, state mapping, calculation management, and result visualization support for the above modules, thereby forming an integrated system for probabilistic modeling, dynamic propagation, simulation solving, and evaluation feedback for the quality characteristics of complex aerospace products.

[0083] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for optimizing the quality characteristics of aviation products based on UQ analysis, characterized in that: It includes: S1: Construct an MBSE system architecture model that integrates probabilistic attributes and establish cross-level constraint relationships for the quality characteristics of aviation products; First, establish the MBSE system architecture model for aviation products. The MBSE system architecture model typically includes a scenario model, a functional model, a logical architecture model, a physical structure model, and a quality characteristic model. Then, the parameters in the MBSE system architecture model are extended with probabilistic attributes. Specifically, based on the source of uncertainty of the parameters, the parameters are divided into four categories: accidental uncertainty parameters, model cognitive uncertainty parameters, deterministic design variables, and task scenario parameters. Finally, a cross-level probabilistic constraint relationship is established between system-level quality characteristics, subsystem functional indicators and underlying physical parameters to form a quality characteristic probability transmission chain, and the cross-level probability transmission relationship is obtained based on the quality characteristic probability transmission chain. S2: Based on the MBSE system architecture model, construct a digital twin model oriented towards quality characteristics to realize dynamic mapping between the physical state of the product and the state of the virtual model; S3: Based on the uncertainty parameters in S1, construct a proxy model based on the design simulation data, and perform a quality characteristic probability propagation analysis based on the design phase information; Model the uncertainty parameters in S1; Based on the parameter dimension and variable coupling characteristics in the cross-level probability transit relationship, an adaptive experimental design method is used to construct a sample set, and a cross-fidelity surrogate model is constructed by combining a high-fidelity physical field analysis model and a low-fidelity fast prediction model; uncertainty propagation analysis is performed using the cross-fidelity surrogate model to calculate the probability distribution characteristics and reliability index of the top-level quality characteristics. S4: Acquire status data during the manufacturing, testing, and service phases, synchronize the virtual model with the physical state through a digital twin model, and dynamically calibrate the cognitive bias of the digital twin model using a Bayesian update mechanism. Acquire state data, including geometric measurement deviations, material performance parameters, and assembly process state data during the manufacturing stage, and map the state data to the digital twin model; Acquire external load, operating environment, and system response timing data during the testing and service phases, and update the digital twin model status synchronously. Using state data as evidence of physical state observation, the model cognitive uncertainty parameters are dynamically corrected using the Bayesian update method; based on the updated posterior distribution of the model cognitive uncertainty parameters, the mass characteristic probability propagation analysis in S3 is re-executed, and then S5 is executed. S5: Perform global sensitivity analysis to obtain key quality impact parameters; S6: Optimize key quality impact parameters and obtain the optimized evaluation results of aviation product quality characteristics.

2. The method for optimizing the quality characteristics of aviation products based on UQ analysis according to claim 1, characterized in that: In S1, the cross-level probability transmission relationship expression obtained from the quality characteristic probability transmission chain is as follows: ； in, Indicates the system-level quality characteristic response. The parameter representing random uncertainty with a probability distribution. This represents the model's cognitive uncertainty parameter. Design variables representing deterministic values, Indicates task scenario parameters. This represents the cross-level logic transfer function defined in the MBSE architecture.

3. The method for optimizing the quality characteristics of aviation products based on UQ analysis according to claim 1, characterized in that: In S3, based on the parameter dimension and variable coupling characteristics in the cross-level probability transitivity, an adaptive experimental design method is used to construct a sample set. A cross-fidelity surrogate model is then constructed by combining a high-fidelity physics analysis model and a low-fidelity fast prediction model. The mathematical expression of the cross-fidelity surrogate model is as follows: ； in, For high-fidelity response prediction values, For low-fidelity response predictions, This is the scale scaling factor. A bias Gaussian process is used to capture the nonlinear differences between high-fidelity and low-fidelity models; The input vector represents the set of key design parameters, process parameters, and environmental uncertainty parameters that affect the system response characteristics; Uncertainty propagation analysis is performed using the aforementioned cross-fidelity proxy model, and the probability distribution characteristics and reliability indices of the top-level quality properties are calculated using the polynomial chaotic expansion (PCE) or Monte Carlo (MCS) methods.

4. The method for optimizing the quality characteristics of aviation products based on UQ analysis according to claim 1, characterized in that: In S4, state data is used as evidence of physical state observation. The Bayesian update method is used to dynamically correct the cognitive uncertainty parameters of the model. The specific formula for the Bayesian update mechanism is as follows: in, To provide physical state observation evidence for the synchronization of digital twin models. For the prior distribution of the model's cognitive uncertainty parameters, For a given model, cognitive uncertainty parameters The likelihood function of the observed data at time, The posterior distribution of the cognitive uncertainty parameters of the dynamically calibrated model; Based on the updated model's understanding of the posterior distribution of uncertainty parameters, the S3 quality characteristic probability propagation analysis is performed again.

5. The method for optimizing the quality characteristics of aviation products based on UQ analysis according to claim 1, characterized in that: Aviation products are hydraulic actuators in flight control systems.

6. The method for optimizing the quality characteristics of aviation products based on UQ analysis according to claim 5, characterized in that: The cross-level probability transitivity relationship in S1 is as follows: ； in, This represents the system-level quality characteristic response, in mission reliability analysis. Flight control system mission reliability In the analysis of instruction tracing capabilities, Take the system-level displacement tracking error or actuation capability margin. This represents a cross-level transfer function formed by connecting contact mechanics models, friction and wear models, leakage models, hydraulic actuator dynamics models, or mission reliability models according to the MBSE architecture; Manufacturing and assembly fluctuation parameters include piston rod diameter deviation, cylinder inner diameter deviation, seal ring size deviation, seal groove size deviation, surface roughness deviation, coaxiality deviation, and assembly compression deviation. This refers to the batch dispersion of material property fluctuation parameters, including the elastic modulus, hardness, Poisson's ratio, and friction coefficient parameters of sealing materials. This includes environmental and mission disturbance parameters, such as fluctuations in hydraulic pressure, ambient temperature, external loads, actuation frequency, and mission load spectrum parameters. and Together they constitute the uncertainty parameters : The contact model cognitive parameters include contact stiffness correction coefficients, boundary constraint correction coefficients, and material constitutive model deviations; The wear model uses cognitive parameters, including wear coefficient, temperature correction coefficient, and degradation rate correction coefficient. The parameters for understanding leakage and hydraulic dynamics models include effective clearance correction factor, leakage model deviation, friction resistance correction factor, and valve control dynamic parameter deviation; , and Together they form the model of cognitive uncertainty parameters ; These are deterministic design variables, including the nominal inner diameter of the cylinder, the nominal diameter of the piston rod, the nominal cross-sectional dimensions of the sealing ring, the nominal sealing clearance, and the design values ​​for surface roughness. These are the parameters for the task scenario, including the task phase, duration, command amplitude, and the number of specified actions.

7. The method for optimizing the quality characteristics of aviation products based on UQ analysis according to claim 5, characterized in that: The digital twin model in S2 includes structural geometric twin, physical behavior twin, and state data twin, which are used to receive real-time state data collected during the manufacturing, testing, and service phases, and to realize the dynamic mapping of the physical product state to the virtual model state. Among them, the structural geometric twin simultaneously preserves the nominal design geometry and the measured geometry of a single part, including the cylinder inner diameter, piston rod diameter, sealing ring cross-sectional dimensions, sealing groove dimensions, sealing gap, surface roughness, coaxiality, and assembly compression, so that the twin model reflects the real geometric state of the single product. The physical behavior twin consists of a contact mechanics model, a friction and wear model, a leakage model, a hydraulic actuator dynamics model, and a task reliability model. Each model reads the corresponding parameters in the cross-level probability transfer relationship and calls an external solver through the analysis model integrated within the MBSE system architecture model or through a dedicated solver interface module. After analysis and solving, the output includes contact pressure, contact deformation, wear depth, leakage amount, output force, response time, displacement tracking error, and task reliability. State data twins are used to receive data from the manufacturing, testing, and service phases, and store them according to the source of the parameters to avoid confusion between measured random fluctuations and model-copied parameters; The digital twin model enables real-time mapping and dynamic updating of quality characteristics based on manufacturing deviations, process execution status, and service environment disturbances.

8. The method for optimizing the quality characteristics of aviation products based on UQ analysis according to claim 5, characterized in that: In S3, for random uncertainty parameters, probability distribution functions are used for modeling. For model cognitive uncertainty parameters, interval probability, prior probability distribution, or evidence structure are used to describe them based on existing validation data. Within the input space composed of the above parameters, the Latin hypercube method is used to generate initial samples, and high-fidelity and low-fidelity models are established according to different responses. High-fidelity models include a nonlinear contact finite element model of the sealing ring, a thermo-fluid-structure interaction leakage model, and a hydraulic actuator system model that includes valve control and pipeline dynamic characteristics. Low-fidelity models include a simplified seal compression model, a rapid wear model, an equivalent gap leakage analytical model, and a reduced-order actuator dynamic model. A cross-fidelity surrogate model is established for each type of quality characteristic. After training the cross-fidelity surrogate model, high-fidelity samples are adaptively increased based on the cross-validation error and the Gaussian process prediction variance until each response meets the preset accuracy. The Monte Carlo method is used to call the validated cross-fidelity surrogate model to complete probability propagation.

9. The method for optimizing the quality characteristics of aviation products based on UQ analysis according to claim 5, characterized in that: In S4, for hydraulic actuators in the flight control system, geometric measurement deviations, material performance parameters, and assembly process status data during the manufacturing stage are acquired, as well as data on pressure, temperature, load, displacement, output force, response time, leakage, wear, and cumulative number of actuations during the testing and service stages. After unit standardization, outlier handling, and time alignment, the actual manufacturing measurements will be used to update the manufacturing and assembly fluctuation parameters in the digital twin model. and material property fluctuation parameters The operating time series is used to update the environmental and task perturbation parameters in the digital twin model. and task scenario parameters The response data serves as physical state observation evidence for calibrating the model's cognitive uncertainty parameters; Based on the sources of the model's cognitive uncertainty parameters, the physical state observation evidence is divided into three groups; Physical observational evidence representing the physical state of contact behavior, including contact pressure, contact deformation, or seal compression; Observational evidence of the physical state of wear and degradation, including wear depth, mass loss, and degradation trajectory as a function of the number of cycles; Evidence of leakage and actuation performance based on physical observations, including leakage amount, output force, response time, and displacement feedback. , and Perform a joint Bayesian update: ； In the formula, To access the cognitive parameters of the model, For wear model cognitive parameters, For understanding parameters of leakage and hydraulic dynamics models, This is a combined vector of all cognitive uncertainty parameters for the model. For the combination of all physical state observation evidence, , and These are the likelihood functions for the cognitive uncertainty parameters of the contact model, the wear model, and the leakage and hydraulic dynamics model, respectively. It is a joint prior distribution; It is a joint posterior distribution.

10. A system for the air product quality characteristic optimization method based on UQ analysis as described in claim 1, characterized in that: It includes: MBSE includes a probabilistic architecture modeling module, a digital twin base module, an uncertainty quantification analysis module, and a dedicated solver interface module; The MBSE probabilistic architecture modeling module is used to construct a probabilistic system architecture model for evaluating the quality characteristics of aerospace products. It includes a hierarchical structure decomposition and model building submodule, a quality characteristic correlation construction submodule, a key quality characteristic parameter definition and description submodule, and an automatic parameter information mapping submodule. The MBSE probabilistic architecture modeling module realizes the cross-level probabilistic expression of the quality characteristics of aerospace products by modeling the quality constraint relationships between the system level, subsystem level, component level, and physical parameter level, and provides a unified parameter relationship foundation for subsequent uncertainty propagation analysis and dedicated solver calls. The digital twin foundation module provides data management, computing support, visualization, and basic functional support, including a database submodule, a general computing submodule, a visualization submodule, and a basic functional submodule. The digital twin base module is used to store and manage product design data, simulation data, manufacturing data, test data, and service status data, and supports dynamic mapping and visualization between physical product status, virtual model status, and quality characteristic evaluation results; The uncertainty quantification analysis module is used to perform uncertainty modeling, propagation analysis and post-processing of the quality characteristics of complex aerospace products. It includes a simulation and external interface call submodule, a random input modeling and sampling submodule, a proxy model construction submodule and a post-processing submodule. The uncertainty quantification analysis module is used to identify uncertainty factors such as manufacturing fluctuations, material dispersion, environmental disturbances and model cognitive biases, construct random input models and generate sampling samples. By calling the simulation model or external solver to obtain the response results, a proxy model is constructed and quality characteristic probability propagation, reliability index calculation, sensitivity analysis and quality risk assessment are performed. The dedicated solver interface module is used to realize data interaction and task scheduling with different types of engineering solvers and computing tools, including cross-level and cross-stage system simulation solvers, finite element solvers, multiphysics coupling solvers, numerical calculation and mathematical analysis solvers, optimization design solvers and parallel computing support modules; This module is used to call the corresponding structural analysis, system simulation, multiphysics coupling, mathematical calculation and optimization design tools according to the analysis requirements of the quality characteristic assessment task, and improves the computational efficiency of high-dimensional uncertainty propagation and multi-condition quality assessment through the parallel computing support module. The hierarchical constraint relationships, key quality characteristic parameters, and parameter mapping relationships output by the MBSE probabilistic architecture modeling module are passed to the uncertainty quantification analysis module. The uncertainty quantification analysis module calls the corresponding engineering solver through the dedicated solver interface module to complete sample response calculation, surrogate model training, and quality risk assessment. The digital twin base module provides data storage, state mapping, computation management and result visualization support for the above modules, thereby forming an integrated system for probabilistic modeling, dynamic propagation, simulation solution and evaluation feedback for the quality characteristics of complex aerospace products.

Images