A method for evaluating performance degradation and reliability of an aircraft door retraction mechanism

Abstract

This invention discloses a method for performance degradation and reliability assessment of aircraft door retraction mechanisms, belonging to the field of aerospace reliability diagnostic technology. It aims to solve the problems of high cost, limited data samples, and low prediction accuracy in existing technologies for full-life-cycle degradation testing of aircraft door retraction mechanisms. The method includes establishing a multibody dynamics model considering hinge clearance and extracting key hinge load boundary conditions; conducting material-level pin-disc friction and wear tests to obtain the basic friction coefficient; constructing an adaptive finite element wear prediction model based on ALE technology; establishing a stochastic process model of the friction coefficient and generating a virtual-real fusion dataset; constructing a performance degradation prediction model based on a physical information neural network, introducing a composite loss function including physical equation residuals and wear monotonicity constraints to achieve small-sample training; and calculating the probability of motion accuracy failure and jamming failure of the mechanism. This method significantly improves the accuracy of reliability assessment while ensuring the physical consistency of the prediction results.

Description

Technical Field

[0001] This invention belongs to the field of aerospace reliability diagnostic technology, specifically relating to a method for evaluating the performance degradation and reliability of an aircraft door retraction mechanism. Background Technology

[0002] The aircraft door retraction mechanism is a crucial component of the landing gear system. Its primary function is to open and close the doors according to a predetermined program during landing gear retraction and extension, maintaining the aircraft's aerodynamic shape and protecting its internal structure. Door retraction mechanisms typically employ multi-link designs, involving complex spatial trajectory transformations and load transfers during their movement. In the overall mission profile of an aircraft, the door closing process presents greater technical challenges and safety risks compared to gravity-assisted opening. This is because, during the closing phase, the driving actuator must not only overcome the mechanism's own weight and frictional drag but also resist aerodynamic loads that vary drastically with flight speed and angle of attack. If the door fails to retract completely or closes inadequately due to mechanism malfunction, it will result in steps or gaps on the aircraft's aerodynamic surfaces, generating significant aerodynamic noise and additional drag.

[0003] Aircraft door retraction mechanisms inevitably face performance degradation during service. Under the alternating high and low temperatures, high humidity and heat, and salt spray corrosion environments of ground parking, taxiing, and flight, the moving parts within the mechanism, especially the hinge pin and bushing interface, are highly susceptible to coupled damage from corrosion and wear. With the passage of service and the increase in the number of operating cycles, this physical degradation leads to a significant decline in the mechanism's macroscopic performance, primarily manifesting as motion accuracy failure and jamming failure. On one hand, continuous wear of the hinge under alternating loads causes a gradual increase in the mating clearance. This clearance is amplified step-by-step in the kinematic chain of the multi-link mechanism, causing the actual movement trajectory of the door end to deviate from the theoretical design value. When the positional deviation in the retracted state exceeds the compensation limit of the sealing strip, sealing failure will occur. On the other hand, with the increase in wear depth, the surface roughness of the contact surface deteriorates. Combined with the third-body effect generated by the accumulation of wear debris, this leads to a significant increase in the internal friction coefficient of the hinge, exhibiting highly nonlinear fluctuation characteristics, which in turn causes an abnormal increase in the overall drag torque during mechanism operation. When the drag torque exceeds the maximum driving force limit that the actuator piston can provide, the mechanism will jam during retraction, leading to mission failure. Therefore, accurately predicting the motion accuracy and drag variation trend of the door retraction mechanism throughout its entire life cycle is crucial to ensuring aircraft reliability.

[0004] However, for performance degradation assessment of such large and complex mechanisms, the engineering and academic communities face the core challenge of effectively addressing the high cost, limited data samples, and low prediction accuracy of full-life-cycle degradation testing for aircraft door retraction mechanisms. Due to the large physical size and complex load conditions of aircraft door retraction mechanisms, conducting accelerated degradation tests on full-life-cycle prototypes requires not only the construction of expensive large-scale environmental simulation test benches but also long test cycles and enormous costs. This results in the acquisition of degradation data from only a very small number of prototypes during actual development, making it difficult to capture the performance dispersion caused by manufacturing and assembly errors using traditional statistical methods. Regarding prediction methods, while traditional physical simulation models based on multibody dynamics have lower computational costs, they are typically based on idealized constant friction coefficients or simplified contact models, making it difficult to realistically reflect the complex nonlinear dynamic responses caused by surface morphology evolution during service, often leading to significant deviations from actual results. In recent years, data-driven methods, represented by deep neural networks, have been applied in fault prediction, but these methods typically rely on massive amounts of training data to fit the input-output relationship. In typical small-sample scenarios like door mechanisms, traditional neural networks are prone to overfitting. Furthermore, due to the lack of physical constraints, purely data-driven models often output predictions that violate physical principles, such as predicting wear decreasing over time or energy non-conservation in the mechanism. Their generalization ability and reliability are severely insufficient. Therefore, how to fully utilize limited material-level and component-level test data, even in the absence of a large number of full-scale experimental samples, to construct a high-precision prediction method that conforms to physical degradation laws and can handle high-dimensional uncertainties is a pressing technical challenge in the field of aerospace reliability engineering.

[0005] Based on this, the present invention proposes a method for evaluating the performance degradation and reliability of aircraft cabin door retraction mechanisms, in order to solve the problems existing in the prior art. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for evaluating the performance degradation and reliability of aircraft door retraction mechanisms, thereby solving the problems of high cost, limited data samples, and low prediction accuracy in the full life cycle degradation test of aircraft door retraction mechanisms in the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for evaluating the performance degradation and reliability of an aircraft cabin door retraction mechanism, comprising: Step S1: Use computational fluid dynamics to obtain the aerodynamic load distribution during the operation of the hatch, establish a multibody dynamics model of the hatch retraction mechanism considering hinge clearance and aerodynamic load, and extract the load boundary conditions of key hinges. Step S2: Based on the material properties of the key hinge, conduct material-level pin-disc friction and wear tests to obtain the basic friction coefficient and wear coefficient under different working conditions; Step S3: Construct an adaptive finite element wear prediction model for the key hinge. Use the basic friction coefficient obtained in step S2 as the model input. At the same time, use the hinge wear test bench to conduct physical wear tests under the load boundary conditions described in step S1. Use the wear depth data obtained from the physical tests to correct and verify the finite element wear prediction model. Step S4: Based on the modified finite element wear prediction model and multibody dynamics model, a stochastic process model considering the time-varying characteristics of the friction coefficient is introduced, and a virtual-real fusion dataset containing the mechanism motion accuracy and resistance degradation data is generated through simulation. Step S5: Construct a performance degradation prediction model based on the Physical Information Neural Network (PINN). The PINN model includes a mutually coupled state prediction network and a degradation rate network, and constructs a composite loss function that includes data error terms, physical dynamic equation residual terms, and monotonicity constraint terms. Step S6: Train the PINN model using the virtual-real fusion dataset generated in step S4, minimize the composite loss function, and combine the trained model with the Monte Carlo simulation method to calculate the motion accuracy failure probability and jamming failure probability of the hatch mechanism.

[0008] In a preferred embodiment of the present invention, in step S1, the aerodynamic load distribution during the operation of the hatch is obtained by computational fluid dynamics, and the hatch and linkage mechanism are set as flexible bodies in the multibody dynamics model, and each connection part is set as a contact pair with gaps.

[0009] In a preferred embodiment of the present invention, step S1, which involves establishing a multibody dynamics model of the hatch retraction mechanism considering hinge clearance and aerodynamic loads, and extracting the load boundary conditions of key hinges, includes: Establish a multi-link mechanism model of the hatch retraction mechanism, and establish a rectangular coordinate system xOy in the simplified kinematic diagram of the mechanism; Wherein, the origin O is located at the fixed rotating hinge G of the hatch, point A is the fixed fulcrum of the actuator, point B is the connecting hinge between the piston rod of the actuator and the linkage mechanism, points C and D are the fixed hinge fulcrum of the linkage mechanism connected to the frame, point E is the connecting node inside the linkage mechanism, point F is the connecting hinge between the linkage mechanism and the hatch, and point G is the fixed rotating hinge of the hatch. Hinges B, E, and F were selected as key research objects, and the dynamic contact force and load torque data of the key hinges under time history were extracted as the load boundary conditions for subsequent physical wear tests.

[0010] In a preferred embodiment of the present invention, in step S2, GCr15 steel is selected as the hinge pin material, brass H62 is selected as the hinge bushing material, and a material-level dry friction test is carried out using a pin-disc friction and wear testing machine. Different test loads and different sliding speeds are set to cover the actual working conditions.

[0011] In a preferred embodiment of the present invention, in step S2, multiple parallel experiments are repeated under each set of experimental conditions, and the arithmetic mean of the experimental results is used as the input parameter for subsequent simulation calculations.

[0012] In a preferred embodiment of the present invention, in step S3, the adaptive finite element wear prediction model employs a node update strategy based on arbitrary Lagrange-Euler (ALE) technology, equating the wear process of nodes to the normal movement of boundary nodes. Based on the calculated wear amount, the ALE algorithm is used to adjust the coordinates of the mesh nodes in the contact area, thereby... ; in, n Let Δ be the normal vector of the contact surface. h This represents the wear amount at each node. The coordinates of the reconstructed mesh nodes. These are the coordinates of the grid nodes before reconstruction.

[0013] In a preferred embodiment of the present invention, the wear amount of each node is calculated using Archard's wear law in each incremental step. The expression of Archard's wear law is: ; in, k The wear coefficient is... P To contact pressure, v For sliding speed, H For material hardness, Δ t The time increment step.

[0014] In a preferred embodiment of the present invention, step S4, which considers the time-varying characteristics of the friction coefficient, is constructed using the Extended Optimal Linear Estimation (EOLE) method, specifically including: Obtain time series data of the friction coefficient and calculate the sample mean and standard deviation; The original friction coefficient data are mapped to a standard Gaussian process using a probability integral transform. Construct the covariance matrix based on the autocorrelation function and standard deviation of the friction coefficient; Eigenvalue decomposition of the covariance matrix yields eigenvalues ​​and eigenvectors; The main eigenvalues ​​and their corresponding eigenvectors are selected to form a reduced-order representation of the friction coefficient stochastic process; the time-varying friction coefficient is decomposed into a weighted combination of a set of uncorrelated standard normal random variables using the EOLE method.

[0015] In a preferred embodiment of the present invention, in step S5, The input vector of the PINN model includes service time. t Key hinge initial clearance parameters ξ 1. ξ 2. ξ 3, and the standard normal random variables x1 to x2 generated by the EOLE method. 15 ; The main body of the network uses a deep neural network (DNN) as its basic architecture and contains multiple hidden layers; The network has a two-branch output structure and is a state prediction network. F (·) is used to directly output predicted values; degradation rate dynamics network. H (·) is used to output auxiliary variables for physical equation calculations; The composite loss function is specifically defined as the weighted sum of the data-driven loss term, the physical information loss term, and the monotonicity constraint loss term.

[0016] In a preferred embodiment of the present invention, in step S6, the reliability assessment includes jamming failure assessment and motion accuracy failure assessment. The function function for jamming failure is defined as the difference between the door running resistance and the maximum driving force of the actuator. When the function function is less than or equal to zero, the mechanism is judged to have jammed failure. The function function for motion accuracy failure is defined as the difference between the deviation of the actual rotation angle and the theoretical rotation angle when the hatch is closed and the allowable threshold. When the function function is less than or equal to zero, the mechanism is determined to have motion accuracy failure. By statistically analyzing the proportion of failed samples in the Monte Carlo sampling, the probability of jamming failure and motion accuracy failure of the hatch opening and closing mechanism during service can be obtained.

[0017] Compared with the prior art, the present invention provides a method for evaluating the performance degradation and reliability of aircraft door retraction mechanisms, which has the following beneficial effects: By constructing a three-level virtual-real fusion testing framework, the degradation tests of large-scale full-size mechanisms, which are difficult to conduct, are decomposed into low-cost material-level pin-plate tests and component-level bench tests. The virtual simulation model is then corrected using physical test data, which significantly reduces the cost and cycle of full life cycle performance degradation tests.

[0018] To address the scarcity of degradation data for aircraft door retraction mechanisms, a physical information neural network is introduced. By embedding the residuals of physical dynamic equations and wear monotonicity constraints into the loss function, physical laws are used as prior knowledge to guide the training of the neural network. This allows the model to converge quickly even with a very small number of training samples, effectively avoiding overfitting and violations of physical common sense that are prone to occur in pure data-driven models with small sample sizes.

[0019] By using the extended optimal linear estimation method to model the friction coefficient, which has highly nonlinear and time-varying characteristics, as a stochastic process, and combining it with Monte Carlo simulation, the reliability level of products in the same batch under the dispersion of physical parameters can be more realistically reflected, providing a scientific basis for condition-based maintenance of the hatch opening and closing mechanism.

[0020] This solves the problems of high cost, small data sample size, and low prediction accuracy in the life cycle degradation test of aircraft cabin door retraction mechanisms in existing technologies. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method for evaluating the performance degradation and reliability of the aircraft cabin door retraction mechanism according to the present invention.

[0022] Figure 2 This is a schematic diagram of the hatch retraction mechanism in an embodiment of the present invention.

[0023] Figure 3 This is a simplified diagram of the door retraction mechanism in an embodiment of the present invention.

[0024] Figure 4 This is a schematic diagram of the adaptive finite element wear of the hinge used in an embodiment of the present invention.

[0025] Figure 5 This is a schematic diagram of the hinge wear test bench used in the component-level hinge wear test in this embodiment of the invention.

[0026] Figure 6 This is a comparison chart of the adaptive wear modeling method used in the embodiments of the present invention and the experimental results.

[0027] Figure 7 This is a schematic diagram of the simulation of the stochastic process of friction coefficient based on the extended optimal linear estimation method in an embodiment of the present invention.

[0028] Figure 8 This is a schematic diagram of the prediction model architecture and loss function construction based on the physical information neural network in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] As per the instruction manual Figure 1 - Appendix Figure 8 As shown in the figure, this invention proposes a method for evaluating the performance degradation and reliability of an aircraft door retraction mechanism. This embodiment takes a door retraction mechanism that controls the opening and closing of the door as an example to illustrate the method of this invention. The aircraft door retraction mechanism mainly includes an actuator, a linkage mechanism, a frame, a door, and a locking ring. The actuator provides the driving force required for the door to retract and extend. One end of the actuator is connected to the frame via a fixed fulcrum, and the other end is connected to the linkage mechanism via a hinge. The frame serves as a support structure fixed to the aircraft fuselage, providing the mounting base for the entire mechanism. The linkage mechanism is a multi-link transmission assembly. One end of the linkage is connected to the piston rod of the actuator via a hinge, and the other end is connected to the door via a hinge. The door is the controlled object, and it is provided with a grid-like reinforcing rib structure to enhance rigidity, used to close the landing gear hatch and maintain the aerodynamic shape of the aircraft; a locking ring is provided at the end of the door. When the door is closed in place, the locking ring re-engages with the locking hook to lock the door.

[0031] This embodiment selects resistance and motion accuracy as two key performance indicators characterizing the health status of the hatch retraction mechanism. During long-term service, hinge wear and deterioration of contact conditions will lead to a continuous increase in the mechanism's resistance. When the resistance exceeds the maximum driving force that the actuator can provide, the mechanism is judged to have jammed and failed. At the same time, increased wear and clearance will affect the accuracy of the hatch closing position. This embodiment uses the deviation between the actual hatch rotation angle and the theoretically designed rotation angle as the motion accuracy indicator. When this deviation exceeds the allowable threshold, the mechanism is judged to have failed in motion accuracy.

[0032] In this embodiment, the function for the door retraction mechanism jamming failure is: ; The effective area of ​​the piston of the mechanism is The maximum pressure that the hydraulic power source can provide is 28 MPa. The maximum driving force that the actuating piston can provide is calculated. It is 24002N.

[0033] The function for failure of the door retraction mechanism's motion accuracy is: ; In the formula, The theoretical angle through which the hatch rotates when the hatch mechanism closes is 54.893°; This refers to the actual angle the hatch rotates when the hatch mechanism closes. The corresponding failure criterion is: when... At that time, the determination mechanism failed due to motion accuracy failure.

[0034] Since the allowable compression of the hatch door sealing strip is 5mm, and the opening and closing process of the hatch can be approximated as a circular arc motion around the center of the fixed hinge, with a radius of motion (i.e., the distance from the hinge center to the contact line of the sealing strip) of 428mm, the corresponding arc length when the sealing strip reaches the allowable compression of 5mm is 5mm. The central angle corresponding to this arc length is... It can be calculated using the arc length formula: ; Therefore, when the actual angular error of the door movement mechanism exceeds 0.669°, the compression of the sealing strip will exceed the allowable value of 5mm, resulting in insufficient compression or overpressure, sealing failure, and thus leakage.

[0035] To accurately predict the degradation trend of the above indicators throughout their entire life cycle, the specific implementation process of the performance degradation and reliability assessment method for the aircraft door retraction mechanism described in this invention is as follows: Figure 1 As shown, it includes the following steps: Step S1: Perform system-level modeling and extract key load boundaries; Step S1.1: In this embodiment, computational fluid dynamics software is first used to numerically simulate the flow field of the aircraft during takeoff and landing to obtain the distribution of unsteady aerodynamic loads on the surface of the door during operation.

[0036] Among them, the aerodynamic load distribution varies with flight speed, angle of attack and hatch opening, and is an important source of external load during the hatch closing process.

[0037] Step S1.2: Subsequently, the pressure field data acting on the outer surface of the hatch calculated by CFD is interpolated and mapped one by one to the surface element nodes corresponding to the flexible body of the hatch in the multibody dynamics model using the radial basis function (RBF) based mesh mapping method.

[0038] Specifically, by writing customized interface scripts, the pressure and shear forces at each time step and node output by the CFD are automatically converted into concentrated forces or distributed loads at the corresponding structural mesh nodes, generating load case files that can be recognized by multibody dynamics software. In this way, the precise application of unsteady aerodynamic loads to the flexible body dynamics model is achieved.

[0039] Step S1.3: Subsequently, a rigid-flexible coupling dynamic model of the hatch retraction mechanism is established in the multibody dynamics simulation software.

[0040] In this model, the hatch and linkage mechanism are treated as flexible bodies to account for the impact of their elastic deformation on motion accuracy. Specifically, modal analysis is performed on the detailed finite element models of the hatch and linkage, and the dominant modes are extracted using the Craig-Bampton modal synthesis method. While ensuring that the dynamic response accuracy error with the fully free-degree-of-freedom model is less than 2%, the first 30 fixed-interface principal modes and all interface constraint modes are extracted. Nodes connected to the actuator and hinge are selected as interface points, generating a flexible body neutral file (.mnf file) containing mass, stiffness matrix, and mode shapes. To handle the contact nonlinearity of the hinge gap, the mesh is refined at the contact interface, and the nonlinear contact force predictor-corrector and parallel sparse matrix solver are enabled in the solver to ensure computational convergence. Then, each connection part is set as a contact pair with a gap to simulate the unavoidable fit gap in actual assembly.

[0041] Specifically, Figure 3 A simplified kinematic diagram of the hatch retraction mechanism in this embodiment is shown. In this embodiment, a rectangular coordinate system xOy is established in the kinematic diagram, with the origin O located at the fixed rotating hinge G of the hatch. Point A in the diagram is the fixed fulcrum of the actuator cylinder, fixedly connected to the frame; point B is the connecting hinge between the actuator cylinder piston rod and the linkage mechanism; points C and D are the fixed hinge fulcrums of the linkage mechanism, connected to the frame; point E is the internal connection node of the linkage mechanism; point F is the connecting hinge between the linkage mechanism and the hatch; and point G is the fixed rotating hinge of the hatch. The diagram also shows the direction of the aerodynamic load acting on the hatch surface.

[0042] This embodiment selects hinges B, E, and F as the key research objects. This selection is based on the fact that, in the complete working cycle of the hatch retraction mechanism, these three hinges have the largest relative rotation angle amplitude compared to other connection nodes, and bear the most significant contact load during the retraction process to overcome aerodynamic drag. They are the most critical parts of the mechanism with the worst working conditions and the greatest susceptibility to wear failure. Based on this multibody dynamics model, a single complete retraction process of the hatch is simulated, and the dynamic contact force and load torque data of hinges B, E, and F over time are extracted. These data will serve as the load boundary conditions for the component-level hinge friction and wear test in subsequent step S3.

[0043] Step S2: Conduct material-level friction and wear tests on the key hinges identified in Step S1 to obtain the basic friction coefficient and wear coefficient of the hinge material pairing under different working conditions.

[0044] In this embodiment, for the key hinge identified in step S1, GCr15 steel is selected as the pin material and H62 brass is selected as the bushing material. The specific mechanical properties of both are shown in Table 1. Since the hardness of GCr15 steel is significantly higher than that of H62 brass, the wear analysis in this embodiment ignores the minor wear of the pin and focuses on the wear behavior of the bushing.

[0045] Table 1: Performance parameters of hinge materials

[0046] Subsequently, dry friction tests at the material level were conducted using a UMT-3 pin-disc friction and wear testing machine. In the tests, the materials of the pin and disc samples were respectively matched to the pin and bushing materials mentioned above. Different test loads and sliding speeds were set to cover the actual working conditions. The test results showed that the friction coefficient exhibited a pattern of large initial fluctuations followed by a more stable trend over time. Based on the test data, a database of basic friction and wear coefficients for this material pair under different working conditions was constructed.

[0047] To reduce the impact of random errors in the experiment, five parallel tests were repeated under each set of experimental conditions, and the arithmetic mean of the test results was used as the input parameter for subsequent simulation calculations. After calculation, the average friction coefficient of the five samples was 0.25204.

[0048] Step S3: Perform component-level virtual-real correction and adaptive modeling; This step constructs an adaptive finite element wear prediction model for the key hinge, using the basic friction coefficient obtained in step S2 as the model input. Simultaneously, physical wear tests are conducted using a hinge wear test bench under the load boundary conditions described in step S1. The wear depth data obtained from the physical tests is used to correct and validate the finite element wear prediction model. Specifically, this includes: Step S3.1: Construct a finite element simulation model of the key hinge and use an adaptive wear modeling method based on arbitrary Lagrange-Euler (ALE) mesh reconstruction technology to simulate the wear evolution process of the hinge during its service life.

[0049] In this model, the wear amount of the contact nodes is calculated based on Arcard's wear law, and the spatial coordinates of the mesh nodes are dynamically adjusted accordingly to update the geometry of the contact surface in real time. Specifically: (1) In terms of the geometric update strategy, the wear process of the node is equivalent to the normal movement of the boundary node. In each increment step, the wear amount Δ of each node is calculated using the Arcard wear law. h The expression for Archard's law of wear is: ; in, k The wear coefficient is... P To contact pressure, v For sliding speed, H For material hardness, Δ t The time increment step.

[0050] (2) Regarding mesh reconstruction, based on the calculated wear amount Δh, the ALE algorithm is used to adjust the mesh node coordinates in the contact area so that: ; in, n The contact surface normal vector, The coordinates of the reconstructed mesh nodes. These are the coordinates of the grid nodes before reconstruction.

[0051] To avoid mesh distortion and computational divergence caused by continuous wear, this embodiment triggers a local mesh re-meshing when the determinant of the Jacobian matrix of any contact region element falls below 30% of its initial value. The re-meshing uses the Advancing Front Method to generate a high-quality local triangular / tetrahedral mesh, and maps the stress-strain field of the old mesh to the new mesh using conserved interpolation. Furthermore, to prevent excessive computation due to frequent re-meshing, a maximum of one re-meshing is triggered per increment step, and the minimum interval between two re-meshings is set to 50 increment steps.

[0052] In this way, the contact geometry is updated without changing the mesh topology, avoiding computational divergence caused by mesh distortion. The wear amount between adjacent incremental steps is determined by calculating the relative position difference of the mesh nodes before and after the update, thus achieving accurate tracking of the wear amount. The simulated adaptive finite element wear distribution cloud map of the hinge is shown below. Figure 4 As shown, where, Figure 4 The left side shows the finite element mesh generation diagram of the pin and bushing fit together. It can be seen that the contact area uses a fine mesh to ensure calculation accuracy. Figure 4 The middle section shows the boundary condition settings diagram, including the normal load application method and displacement constraint conditions; Figure 4 The right side shows a magnified view of the wear depth cloud map of the wear area. It can be clearly seen from the cloud map that the maximum wear occurs at the center of the contact area and gradually decreases towards the edge.

[0053] Step S3.2: Using, for example Figure 5 The rotating hinge wear test bench shown was used to conduct component-level physical tests to verify the above simulation model.

[0054] Specifically: The hinge specimen is mounted in the test fixture, and accelerated wear testing is performed by applying the dynamic load spectrum extracted in step S1. During the test, a high-precision laser displacement sensor is used to periodically monitor the relative radial displacement between the pin and the bushing, which serves as characterization data for the actual wear depth. The wear depth data obtained from the physical test is then compared and analyzed with the finite element simulation results. The comparison results are as follows: Figure 6 As shown, where, Figure 6 The horizontal axis represents the number of actuations, and the vertical axis represents the wear depth. The figure also plots the experimental data curves and the adaptive finite element model prediction curves.

[0055] The results show that the simulated prediction curves agree well with the physical test data, verifying the effectiveness of the finite element wear prediction model. Based on this comparison, the local wear coefficient k in the finite element model is reverse-calibrated and corrected using the least squares method, thus finally obtaining a high-fidelity component-level wear prediction model verified by physical tests. This model can accurately predict the wear evolution of hinges under different working conditions with relatively low computational resource consumption.

[0056] Step S4: Generation of uncertain samples based on EOLE; This step is based on the modified finite element wear prediction model and multibody dynamics model. It introduces a stochastic process model that considers the time-varying characteristics of the friction coefficient and generates a virtual-real fusion dataset containing the motion accuracy and resistance degradation data of the mechanism through simulation.

[0057] To achieve coupled simulation of wear and mechanism performance degradation, this embodiment employs a sequential coupling strategy. Whenever the finite element wear model predicts that the cumulative wear depth of hinges B, E, and F reaches a preset update threshold, the wear simulation is paused. The wear profile features of the hinge contact surfaces at this point are extracted, including the maximum depth and wear width. Based on Hertz contact theory and tribological empirical formulas, the current time-varying friction coefficient and equivalent radial clearance are recalculated. Subsequently, these updated hinge parameters are written back to the multibody dynamics model, updating the corresponding contact pairs and friction models. A complete extension / retraction cycle of dynamics simulation is then run again to obtain the resistance and motion accuracy under the current degradation state. This iterative closed loop is repeated until the preset total number of cycles is reached. Specifically: Considering the significant stochastic time-varying characteristics of the friction coefficient in the service environment, this embodiment employs the Extended Optimal Linear Estimation (EOLE) method to construct a stochastic process model. Specifically, it includes: Step S4.1: First, based on the friction coefficient time series data obtained in steps S2 and S3, as shown in Table 2, the sample mean of the friction coefficient is calculated as follows: The standard deviation is .

[0058] Table 2: Time series data of friction coefficient

[0059] Given that the original friction coefficient data exhibits a significant non-Gaussian distribution, this embodiment uses probability integral transformation to map it into a standard Gaussian process to facilitate subsequent mathematical processing.

[0060] Step S4.2: Next, the stochastic process is discretized using the Extended Optimal Linear Estimation (EOLE) method. The specific steps are as follows: Step S4.2.1: Construct the covariance matrix based on the autocorrelation function and standard deviation of the friction coefficient; Step S4.2.2: Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors; Step S4.2.3: Select the main eigenvalues ​​and their corresponding eigenvectors to form a reduced-order representation of the random process of friction coefficient.

[0061] Specifically, in this embodiment, retaining 15 key feature terms is sufficient to meet the engineering accuracy requirements. Therefore, 15 standard normal random variables are used to characterize the stochastic process of the friction coefficient. The time-varying friction coefficient is decomposed into a weighted combination of a set of uncorrelated standard normal random variables using the EOLE method, thereby effectively characterizing the input uncertainty caused by changes in surface roughness and the third-body effect.

[0062] This embodiment compares and analyzes three autocorrelation function models: the exponential model, the squared exponential model, and the exponential model with an oscillating term. The fitting results of the autocorrelation function are as follows: Figure 7 As shown. Figure 7 The x-axis represents the cyclic hysteresis τ, and the y-axis represents the autocorrelation function R(τ). The figure compares the fitting effects of the experimental data with the three models. Based on the comparison of fitting errors, the oscillating model with the best consistency with the experimental data was selected as the autocorrelation function of the friction coefficient stochastic process to accurately capture the oscillating decay characteristics of the friction coefficient.

[0063] This allows us to obtain a stochastic process model for characterizing the stochastic time-varying properties of the friction coefficient.

[0064] Step S4.3: Finally, considering load fluctuations, time-varying friction coefficients, and initial hinge gap dispersion, a large-scale simulation is performed under various uncertainties to generate a virtual-real fusion sample dataset for different operating conditions. This dataset contains information on the resistance changes and motion accuracy degradation of the hatch retraction mechanism under different uncertainties, providing a sufficient data foundation for training the surrogate model in the subsequent step S5.

[0065] Step S5: Construct a performance degradation prediction model based on a physical information neural network; This step, based on the virtual-real fusion dataset generated in step S4, constructs a physical information neural network (PINN) performance degradation prediction model, the structure of which is as follows: Figure 8 As shown, the model comprises a coupled state prediction network and a degradation rate network, and constructs a composite loss function that includes data error terms, physical dynamic equation residual terms, and monotonicity constraint terms.

[0066] Figure 8 This illustrates the PINN prediction model architecture and loss function construction method used in this embodiment. The model's input vector includes service time t and the initial clearance parameter of the critical hinge. ξ 1. ξ 2. ξ 3, and 15 standard normal random variables generated by the EOLE method. x 1 to x 15 The network primarily uses a deep neural network (DNN) as its basic architecture, containing multiple hidden layers, each employing a non-linear activation function for feature extraction. The network has a dual-branch output structure: a state prediction network... F (·) Used for directly outputting predicted values ​​such as drag and angular deviation, degradation rate dynamics network H (·) is used to output auxiliary variables for physical equation calculations.

[0067] During model training, the sample fitting error, degradation process consistency constraint, and monotonicity constraint are all used as the loss function. This approach enables the model to not only fit the simulated samples but also satisfy the basic degradation law of increased resistance and decreased motion accuracy after wear accumulation. The composite loss function is specifically defined as the weighted sum of the data-driven loss term, the physical information loss term, and the monotonicity constraint loss term.

[0068] The data-driven loss term compares the network output with the true value to calculate the error, ensuring the model's fitting accuracy to the training samples. Its expression is the mean squared error of the difference between the predicted and true values ​​of the training samples. The physical information loss term uses automatic differentiation to calculate the derivative and combines it with the physical dynamics equations of mechanism degradation to map and calculate the residuals of the equations, ensuring that the network output satisfies the physical conservation laws. The monotonicity constraint loss term addresses the irreversible nature of the wear process by penalizing the predicted non-monotonic degradation trend. This is achieved by adding the square of the difference in degradation at adjacent time points as a penalty term to the loss function, forcing the model output to conform to the physical law of monotonically increasing wear.

[0069] The PINN model exhibits significant advantages over traditional data-driven neural networks under limited sample conditions. Traditional neural networks are prone to overfitting when the sample size is insufficient, and the lack of physical constraints can lead to predictions that violate physical principles, such as predicting wear as decreasing over time. This invention, however, embeds the residuals of physical dynamic equations and wear monotonicity constraints into the loss function, using physical laws as prior knowledge to guide neural network training. This allows the model to converge quickly and output physically consistent predictions even with a very small number of training samples.

[0070] Step S6: Reliability assessment; This step uses the virtual-real fusion dataset generated in step S4 to train the PINN model, minimizes the composite loss function, and combines the trained model with the Monte Carlo simulation method to calculate the motion accuracy failure probability and jamming failure probability of the hatch mechanism.

[0071] After training, the PINN model is used as a fast surrogate model to replace time-consuming multibody dynamics simulations, and reliability assessment is conducted using Monte Carlo sampling. The random variables shown in Table 3 are used as inputs, including the initial clearance parameters of key hinges. ξ 1. ξ 2. ξ 3. Aerodynamic load parameters F and 15 random variables generated by EOLE x 1 to x 15 Calculate the resistance and closing angle deviation for different samples.

[0072] Table 3: Distribution Parameters of Influencing Factors

[0073] In this embodiment, the function function for the jamming failure of the hatch retraction mechanism is defined as follows: G 1= F 1- F 0; in, F 1 represents the resistance to hatch operation. F 0 represents the maximum driving force of the actuator. The effective area of ​​the actuator piston is A = π × (0.023) 2 - 0.016 2 ) = 8.57 × 10 -4 m 2 The maximum pressure that the hydraulic power source can provide is 28 MPa. The maximum driving force that the actuating piston can provide is calculated. F 0 is 24002N. When the function... G When 1 is less than or equal to zero, the determination mechanism fails due to jamming.

[0074] The function function for the failure of the door retraction mechanism in terms of motion accuracy is defined as follows: G 2= δ 0- δ 1; in, δ 0 represents the allowable threshold for motion accuracy. δ 1 represents the angular deviation when the hatch is closed.

[0075] Theoretical angle of rotation of the hatch when the hatch mechanism closes θ 0 is 54.893°, θ 1 represents the actual angle the hatch rotates when the hatch mechanism closes. The allowable compression of the sealing strip is 5mm. The opening and closing process of the hatch can be approximated as a circular arc motion around the center of the fixed hinge, with a radius of 428mm. When the sealing strip reaches the allowable compression of 5mm, the corresponding arc length is 5mm. The corresponding central angle, i.e., the allowable threshold for motion accuracy, is calculated according to the arc length formula. δ 0 is approximately 0.669°. Therefore, when the actual angular error of the door movement mechanism exceeds 0.669°, the mechanism is deemed to have suffered a motion accuracy failure.

[0076] By conducting extensive Monte Carlo sampling within the distribution range of random variables, a pre-trained PINN surrogate model is used to quickly calculate the drag and angular deviation corresponding to each sample. The percentage of samples where drag and angular deviation exceed thresholds is then statistically analyzed, thereby obtaining the probability of jamming failure and motion accuracy failure of the hatch retraction mechanism during service. This method significantly reduces the computational cost of reliability assessment while maintaining prediction accuracy, providing a scientific basis for condition-based maintenance and life management of hatch retraction mechanisms.

[0077] In summary, this invention constructs a three-level virtual-real fusion testing framework integrating system-level simulation, component-level correction, and material-level fundamental data. By combining a physical information neural network (PEN) with Monte Carlo simulation, it achieves accurate prediction and reliability assessment of the full life-cycle performance degradation of aircraft door retraction mechanisms. The virtual-real fusion testing framework effectively solves the problems of high cost and limited data samples in full-scale physical prototype degradation testing; the PSN addresses the issues of prediction accuracy and physical consistency under small sample conditions by embedding physical law constraints; and the combination of EOLE stochastic process modeling and Monte Carlo simulation enables accurate quantification of complex stochastic degradation processes.

[0078] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for performance degradation and reliability assessment of an aircraft door retraction mechanism, characterized in that, include: Step S1: Use computational fluid dynamics to obtain the aerodynamic load distribution during the operation of the hatch, establish a multibody dynamics model of the hatch retraction mechanism considering hinge clearance and aerodynamic load, and extract the load boundary conditions of key hinges. Step S2: Based on the material properties of the key hinge, conduct material-level pin-disc friction and wear tests to obtain the basic friction coefficient and wear coefficient under different working conditions; Step S3: Construct an adaptive finite element wear prediction model for the key hinge. Use the basic friction coefficient obtained in step S2 as the model input. At the same time, use the hinge wear test bench to conduct physical wear tests under the load boundary conditions described in step S1. Use the wear depth data obtained from the physical tests to correct and verify the finite element wear prediction model. Step S4: Based on the modified finite element wear prediction model and multibody dynamics model, a stochastic process model considering the time-varying characteristics of the friction coefficient is introduced, and a virtual-real fusion dataset containing the mechanism motion accuracy and resistance degradation data is generated through simulation. Step S5: Construct a performance degradation prediction model based on the Physical Information Neural Network (PINN). The PINN model includes a mutually coupled state prediction network and a degradation rate network, and constructs a composite loss function that includes data error terms, physical dynamic equation residual terms, and monotonicity constraint terms. Step S6: Use the virtual-real fusion dataset generated in step S4 to train the PINN model, minimize the composite loss function, and combine the trained model with the Monte Carlo simulation method to calculate the motion accuracy failure probability and jamming failure probability of the hatch mechanism.

2. The method of claim 1, wherein the method further comprises: In step S1, the aerodynamic load distribution during the operation of the hatch is obtained using computational fluid dynamics. In the multibody dynamics model, the hatch and linkage mechanism are set as flexible bodies, and each connection part is set as a contact pair with gaps.

3. The method of claim 1, wherein In step S1, the process of establishing a multibody dynamics model of the hatch retraction mechanism considering hinge clearance and aerodynamic loads, and extracting the load boundary conditions of key hinges, includes: Establish a multi-link mechanism model of the hatch retraction and extension mechanism, and establish a rectangular coordinate system xOy in the simplified kinematic diagram of the mechanism; In the rectangular coordinate system xOy, the origin O is located at the fixed rotating hinge G of the hatch, point A is the fixed fulcrum of the actuator, point B is the connecting hinge between the piston rod of the actuator and the linkage mechanism, points C and D are the fixed hinge fulcrum of the linkage mechanism connected to the frame, point E is the connecting node inside the linkage mechanism, point F is the connecting hinge between the linkage mechanism and the hatch, and point G is the fixed rotating hinge of the hatch. Hinges B, E, and F were selected as key research objects, and the dynamic contact force and load torque data of the key hinges under time history were extracted as the load boundary conditions for subsequent physical wear tests.

4. A method of performance degradation and reliability assessment of an aircraft door retraction mechanism as defined in claim 1, characterized in that, In step S2, GCr15 steel is selected as the hinge pin material, and H62 brass is selected as the hinge bushing material. A material-level dry friction test is carried out using a pin-disc friction and wear tester. Different test loads and different sliding speeds are set to cover the actual working conditions.

5. A method of performance degradation and reliability assessment of an aircraft door retraction mechanism as claimed in claim 4, wherein, In step S2, multiple parallel experiments are repeated under each set of experimental conditions, and the arithmetic mean of the experimental results is used as the input parameter for subsequent simulation calculations.

6. A method of performance degradation and reliability assessment of an aircraft door retraction mechanism as defined in claim 1, wherein, In step S3, the adaptive finite element wear prediction model adopts a node update strategy based on the arbitrary Lagrange-Euler (ALE) technique, which equates the wear process of nodes to the normal movement of boundary nodes. Based on the calculated wear amount, the ALE algorithm is used to adjust the mesh node coordinates in the contact area, so that... ; wherein, n is the contact surface normal vector, Δ h is the wear amount of each node, is the reconstructed mesh node coordinate, is the mesh node coordinate before reconstruction.

7. A method of performance degradation and reliability assessment of an aircraft door retraction mechanism as claimed in claim 6, wherein, In each increment step, the wear amount of each node is calculated using Archard's wear law. The expression for Archard's wear law is: ； wherein, k is a wear coefficient, P is a contact pressure, v is a sliding speed, H is a material hardness, Δ t is a time increment step.

8. A method of assessing performance degradation and reliability of an aircraft door retraction and extension mechanism as recited in claim 1, wherein, In step S4, the stochastic process model considering the time-varying characteristics of the friction coefficient is constructed using the Extended Optimal Linear Estimation (EOLE) method, specifically including: Obtain time series data of the friction coefficient and calculate the sample mean and standard deviation; The original friction coefficient data are mapped to a standard Gaussian process using a probability integral transform. Construct the covariance matrix based on the autocorrelation function and standard deviation of the friction coefficient; Eigenvalue decomposition of the covariance matrix yields eigenvalues ​​and eigenvectors; The main eigenvalues ​​and their corresponding eigenvectors are selected to form a reduced-order representation of the friction coefficient stochastic process; the time-varying friction coefficient is decomposed into a weighted combination of a set of uncorrelated standard normal random variables using the EOLE method.

9. A method of performance degradation and reliability assessment of an aircraft door retraction mechanism as defined in claim 1, wherein, In step S5, The input vector of the PINN model includes service time. t Key hinge initial clearance parameters ξ 1. ξ 2. ξ 3, and the standard normal random variables x1 to x2 generated by the EOLE method. 15 ; The main body of the network uses a deep neural network (DNN) as its basic architecture and contains multiple hidden layers; The network has a two-branch output structure and is a state prediction network. F (·) is used to directly output predicted values; degradation rate dynamics network. H (·) is used to output auxiliary variables for physical equation calculations; The composite loss function is specifically defined as the weighted sum of the data-driven loss term, the physical information loss term, and the monotonicity constraint loss term.

10. A method of performance degradation and reliability assessment of an aircraft door retraction and extension mechanism as defined in claim 1, characterized in that, In step S6, the reliability assessment includes jamming failure assessment and motion accuracy failure assessment; The function function for jamming failure is defined as the difference between the door running resistance and the maximum driving force of the actuator. When the function function is less than or equal to zero, the mechanism is judged to have jammed failure. The function function for motion accuracy failure is defined as the difference between the deviation of the actual rotation angle and the theoretical rotation angle when the hatch is closed and the allowable threshold. When the function function is less than or equal to zero, the mechanism is determined to have motion accuracy failure. By statistically analyzing the proportion of failed samples in the Monte Carlo sampling, the probability of jamming failure and motion accuracy failure of the hatch opening and closing mechanism during service can be obtained.

Images