Aviation O-shaped sealing ring service life prediction method and structure optimization method
By combining the PINN model with physical degradation laws and data-driven methods, the problem of insufficient O-ring life prediction accuracy was solved, achieving efficient life prediction and structural optimization, and improving the design reliability and service life of the sealing structure.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-03
- Publication Date
- 2026-03-27
AI Technical Summary
In the existing technology, the life prediction method of O-rings has insufficient prediction accuracy under the influence of multiple factors, is highly dependent, and has poor engineering applicability, making it difficult to accurately describe its degradation law in complex systems.
A PINN-integrated data-driven and physical prior method was adopted to construct an O-ring lifetime prediction model. Through accelerated aging tests and data-driven regression analysis, combined with physical degradation models and neural networks, lifetime prediction and structural optimization were achieved.
It improves the accuracy and engineering applicability of O-ring life prediction, reduces reliance on test data, and enables efficient design and extended lifespan of sealing structures.
Smart Images
Figure CN121744531A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aerospace component structure optimization, specifically to a method for predicting the lifespan of an aerospace O-ring seal and a method for structural optimization. Background Technology
[0002] O-rings are circular sealing rings. As a critical basic component, they are typically made of thermoplastic elastomers, rubber, silicone, and other materials, and are widely used in various static or dynamic sealing structures across industries such as aerospace, petrochemicals, power equipment, and rail transportation. Although O-rings are low-cost consumable parts, they play a vital role in preventing media leakage, maintaining system stability, and ensuring the safe operation of equipment. Especially in engineering applications such as aerospace, where high reliability, long lifespan, and complex operating conditions are increasingly demanding, the performance of O-rings has a significant impact on the overall reliability and service life of the system.
[0003] In actual operation, the performance and lifespan of O-rings are significantly affected by key factors such as temperature environment and sealing groove structure. Temperature changes directly alter the mechanical and elastic properties of the O-ring material. When the temperature rises, the material may experience thermal expansion or performance degradation, leading to increased sealing gap and reduced contact stress, thereby weakening the sealing effect and accelerating the aging process. The sealing groove depth, as one of the core design parameters of the sealing structure, affects the O-ring's compressibility, contact state, and stress distribution. An unreasonable groove depth design may result in insufficient sealing or excessive compression, leading to premature failure. Therefore, accurately assessing the impact of temperature and groove depth on O-ring lifespan during the sealing structure design phase, and selecting design parameters based on lifespan prediction results, is of significant engineering importance. This not only effectively extends the service life of O-rings and reduces maintenance costs but also improves the overall system's operational reliability and safety.
[0004] Currently, the main methods for predicting the lifespan of O-rings include physical modeling and data-driven methods.
[0005] The physical model method takes the failure modes and mechanisms of O-rings as its starting point and has high interpretability and reliability. However, this method requires a clear understanding of the internal working principle and failure mechanism of O-rings in order to conduct reliability and life assessment. When faced with complex systems or situations with multiple coupled factors, this method is difficult to establish a complete, accurate and comprehensive failure degradation model. Therefore, it cannot accurately describe the degradation law of O-rings when considering multiple factors.
[0006] Data-driven methods have been widely applied in O-ring lifetime prediction and reliability assessment. Mathematical statistics methods primarily rely on historical data of system state changes to predict current system state changes by establishing different degradation models (such as Dakin models and variable reduction models). However, this method is highly dependent on the model, has limited application in nonlinear scenarios, and its lifetime prediction capability under complex variable conditions is somewhat limited, requiring further in-depth research. Summary of the Invention
[0007] To address the shortcomings of existing technologies, the present invention aims to provide a life prediction method and structural optimization method for aerospace O-rings. This method solves the limitations of existing technologies, such as single physical models or traditional data-driven methods in the field of life prediction, insufficient prediction accuracy under multiple influencing factors, strong dependence on experiments, and poor engineering applicability. The present invention adopts the PINN method, which integrates data-driven and physical priors, to provide a unified modeling approach for O-ring life prediction, structural optimization, and engineering applications.
[0008] Specifically, the present invention provides a method for predicting the lifespan of an aerospace O-ring seal, which includes the following steps: S1: Conduct accelerated aging tests: Conduct O-ring life tests, measure the dimensional parameters of O-rings at different aging times and calculate their compression set, obtain O-ring test data under various accelerated temperatures and groove depths, and construct a basic degradation sample set; S2: Constructing an O-ring performance degradation model: Based on the dynamic curve model, construct an O-ring performance degradation model at various acceleration temperatures. Using data-driven regression analysis, calibrate the parameters of the O-ring performance degradation model at various acceleration temperatures according to the basic degradation sample set constructed in step S1, and obtain the O-ring performance degradation model at the actual operating temperature. S3: Life test data expansion: Based on the performance degradation models at various accelerated temperatures constructed in step S2, the basic degradation sample set constructed in step S1 is expanded using a data-driven method to obtain an effective supplementary life test dataset under limited test conditions, thus obtaining a database for machine learning. S4: Construct and train an O-ring lifetime prediction model based on the PINN model, including the following sub-steps: S41: Construct an O-ring lifetime prediction model based on the PINN model. The neural network takes acceleration temperature, groove depth and compression permanent deformation retention rate as input parameters and lifetime as output. Set the number of layers and the number of neurons in each layer of the hidden layer structure, and adjust the parameters of the hidden layer to optimize the fitting ability of the nonlinear relationship between the input parameters and the output. S42: The data in the database used in the machine learning in step S3 enters the neural network from the input layer, is calculated layer by layer through each hidden layer and passed to the output layer to form the prediction result; S43: Calculate the loss function using the backpropagation algorithm combined with gradient descent, and update the weights and biases; S44: Based on the O-ring performance degradation model at actual operating temperature obtained in step S2, the lifetime time is introduced into the O-ring lifetime prediction model as a physical law in the form of a function, and the root mean square error is used as the loss function. The physical law is integrated into the loss function in an explicit construction method: ; in, is the compression set retention rate of the O-ring; T is the acceleration temperature; h1 is the groove depth; is the physical weight, used to adjust the degree to which the model conforms to physical laws; N is the lifetime of the O-ring. This represents the true lifetime value of the i-th sample. This represents the predicted lifetime value output by the neural network for the i-th sample; n represents the number of samples in the dataset. For physical degradation models; By adjusting various hyperparameters, an iterative training method is used to obtain a lifetime prediction model that integrates data fitting and physical constraints.
[0009] Further: In step S1, the method for calculating the compressive permanent deformation rate of the O-ring is as follows: ; In the formula, h0 is the compression permanent deformation rate of the O-ring; h1 is the original cross-sectional diameter of the O-ring; h2 is the groove depth of the O-ring; h2 is the radial cross-sectional height of the O-ring after deformation.
[0010] Further: In step S2, the O-ring performance degradation model at the actual operating temperature is as follows: ; In the formula, B and α are experimental constants; K is the aging rate constant; and t is the aging time.
[0011] Further: The method for expanding the basic degradation sample set in step S3 is as follows: the parameters of the O-ring performance degradation model at actual operating temperature are calibrated using data-driven regression analysis and the basic degradation sample set. Using the calibrated O-ring performance degradation model, the existing experimental data are mapped to new temperature and groove depth conditions to generate equivalent degradation curves under untested conditions.
[0012] Furthermore: the neural network constructed in step S41 calculates the weighted input: ; In the formula, x is the input parameter; w is the weight; and b is the bias. The output parameters are obtained after being transformed by a transfer function.
[0013] Furthermore, step S43 specifically includes the following sub-steps: S431: Error Calculation: Calculate the errors of the output layer and the hidden layer respectively; S432: Weight Update: Update the weights and biases of the neural network based on the gradient of the error function; S433: Method Optimization: Optimize the constructed neural network using one or more of the SGD, Adam, and Levenberg-Marquardt algorithms.
[0014] Further: The error of the output layer in step S431 is: ; In the formula, L is the loss function constructed in step S44. For the first Layer error term, In a neural network, the first... The linear combination of layers is the input before the activation function is applied.
[0015] Further: The hidden layer error propagation formula in step S431 is: ; in, This represents element-wise multiplication; The derivative of the transfer function; error It is used as the gradient of the loss function for the next neuron in the backpropagation direction, and is passed sequentially until it reaches the input layer.
[0016] This invention also provides a structural optimization method for O-rings used in aviation. Based on the aforementioned method for predicting the lifespan of O-rings used in aviation, and according to the constructed O-ring lifespan prediction model, the optimal O-ring groove depth parameter is obtained with the aging lifespan of the O-ring as the optimization objective, which is used to support the selection of sealing structures.
[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention provides an O-ring life prediction model based on a physical constraint neural network. By introducing a sealing ring performance degradation model as a physical constraint into the PINN framework, the model can simultaneously possess physical interpretability and data generalization ability, achieving a deep integration of degradation mechanism and data-driven approach. It can provide more realistic O-ring life prediction results in engineering applications, and provide an effective tool for accurate and efficient sealing structure design optimization and life and performance prediction.
[0018] 2. This invention constructs a physical consistency expansion method for multi-condition degradation data. Under finite life test sample conditions, it expands the basic dataset using degradation kinetics. Using the calibrated O-ring performance degradation model, it maps existing test data to new temperature and tank depth conditions, generating equivalent degradation curves under untested conditions. This effectively improves the sample diversity of model training, overcomes the problem of low accuracy of life prediction results due to insufficient test data, and significantly reduces the dependence of life prediction on a large amount of test data.
[0019] 3. This invention utilizes the gradient backpropagation characteristics of the PINN model to establish a life-sensitivity backpropagation optimization mechanism. This mechanism enables sensitivity analysis and rapid optimization of sealing structure parameters, including groove depth and compression set, thereby reducing testing costs and providing quantitative guidance for the structural optimization design and service life extension of O-rings used in aviation. Attached Figure Description
[0020] Figure 1 This is an overall flowchart of the life prediction method for aviation O-ring seals disclosed in this invention. Figure 2 This is a detailed flowchart of each step in the life prediction method for aviation O-ring seals disclosed in this invention; Figure 3 This is a diagram of the neural network model constructed in step S4 of the life prediction method for aviation O-rings disclosed in this invention. Detailed Implementation
[0021] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
[0022] Combination Figure 1 and Figure 2 As shown, this invention provides a method for predicting the lifespan of an aerospace O-ring seal, comprising the following steps: S1: Accelerated thermal aging test was adopted to carry out O-ring life test under different temperature and groove depth conditions. The key dimensions of the O-ring were measured at different aging times, and its compression permanent deformation rate was calculated according to the following formula. O-ring life test data under different accelerated temperature and groove depth conditions were obtained to construct a basic degradation sample set. ; In the formula, h0 is the compression permanent deformation rate of the O-ring; h1 is the original cross-sectional diameter of the O-ring; h2 is the groove depth of the O-ring; h2 is the radial cross-sectional height of the O-ring after deformation.
[0023] The accelerated aging test process is as follows: First, the two main factors affecting O-ring life are determined to be temperature and the depth of the sealing structure groove. The influence level of temperature is 5, and the influence level of the sealing structure depth is 3. Specifically, the temperature T is (90℃, 110℃, 120℃, 130℃, 140℃); and the sealing structure groove depth h1 is (6mm, 7mm, 8mm). Based on the accelerated thermal aging test, O-ring aging life tests are then conducted under 15 different influencing factors.
[0024] During the aging test, samples were taken at the planned time, and the key dimensions of the O-rings were measured according to the test specifications. Their compression set was calculated, and O-ring life test data were obtained under different acceleration temperatures and groove depths. This resulted in 15 tables and 120 sets of test data under different conditions, constructing a basic degradation sample set for the O-rings. It is worth noting that, for the planned time and different temperature conditions, the compression value at the end of the final test should be as close to the critical value as possible.
[0025] Table 1. Experimental design table for a temperature of 90℃ and a tank depth of 6mm: S2: Based on the kinetic curve model, a degradation trajectory model of the O-ring's performance under accelerated temperature as a function of aging time was constructed (2). The parameters of the O-ring performance degradation model were calibrated using iterative analysis and the least squares method with the basic degradation sample set to obtain the O-ring performance degradation curves at different temperatures. Using the Arrhenius formula (3), ln(B) and the aging rate constant K were calibrated by the least squares method, and the O-ring performance degradation model at the actual operating temperature was obtained by regression.
[0026] (2) In the formula, α is the experimental constant; K is the aging rate constant; t is the aging time (d or min); The compression permanent deformation rate of the seal.
[0027] The parameters of the O-ring performance degradation model were calibrated using data-driven regression analysis and a basic degradation sample set. The α value for different sealing structure groove depths was calibrated using a successive approximation method.
[0028] Within a certain temperature range, the aging rate constant K follows the Arrhenius model and is related to the thermodynamic temperature T. (3) In the formula, A is the characteristic factor (d-1 or min-1); E is the activation energy (J / mol); and R is the molar gas constant (R=8.314J / (mol•K)).
[0029] S3: Based on the performance degradation models obtained above at different temperatures, the basic degradation sample set is expanded using a data-driven approach. That is, using the calibrated degradation rate model, the existing experimental data is mapped to new temperature and tank depth conditions to generate "equivalent degradation curves" under untested conditions. This yields an effective supplementary lifetime dataset under limited experimental conditions, providing a database for machine learning.
[0030] S4: Construct and train an O-ring lifetime prediction model based on the PINN model, including the following sub-steps: S41: Construct an O-ring lifetime prediction model based on the PINN model, and build a neural network, such as... Figure 3 As shown. The number of neurons in the input and output layers corresponds to the number of input parameters and output results, respectively. The number of hidden layers and the number of neurons in each layer are adjustable. Training is performed using the backpropagation algorithm, which includes forward propagation, backpropagation, and weight updates. Physical laws are embedded in the loss function, using root mean square error (RMSE) as the loss function. By adjusting the hyperparameters, the final training result is obtained, achieving accurate prediction of O-ring lifespan.
[0031] The basic unit of a neural network is a neuron. Each neuron receives an input vector x and calculates a weighted input using weights w and biases b. The output y is obtained after being transformed by the transfer function f.
[0032] Each neuron in each layer connects to neurons in the next layer via weights. Connected. Assume the first... There are layers The _ neuron, the _ ... There are layers If there are 10 neurons, then the set of weights between the two layers is 10. weight matrix Meanwhile, the data is normalized using a transfer function, compressing the output values to a certain range. By continuously adjusting the parameters of the hidden layer, the fitting and prediction of the complex nonlinear relationship between the input and output can be further optimized.
[0033] The number of neurons in the input layer corresponds to the model input parameters, including temperature, depth of the sealing structure groove, and compression set retention rate.
[0034] Number of output layer neurons: corresponds to the prediction target; in this embodiment, it is the lifetime of the O-ring.
[0035] S42: Data enters the network from the input layer, is calculated layer by layer through each hidden layer, and is then passed to the output layer to form the prediction result. Layer and First The relationship between neurons in a layer can be represented as: ; ; in, It is the first Weighted input of the layer, It is the first Layer activation output, It is a transfer function (such as Sigmoid or ReLU). Through layer-by-layer transfer, the final output of the network is obtained. .
[0036] S43: The loss function is calculated using the backpropagation algorithm combined with gradient descent, and the weights and biases are updated. This mainly includes: S431: Error calculation.
[0037] For the output layer, the error can be expressed as: .
[0038] For the hidden layer, the error propagation formula is: ; in, This represents element-wise multiplication. It is the derivative of the transfer function. Error. It is used as the gradient of the loss function for the next neuron in the backpropagation direction, and is passed sequentially until it reaches the input layer.
[0039] S432: Weight update.
[0040] Update the network weights based on the gradient of the error function. and bias The updated formula is: ; ; in, This is the learning rate, typically set between 0.01 and 0.001, and L is the loss function constructed in step S44. For the first Layer error term, In a neural network, the first... The linear combination of layers is the input before the activation function is applied.
[0041] S433: Method optimization.
[0042] In actual training, optimization algorithms such as SGD (Stochastic Gradient Descent), Adam (Adaptive Moment Estimation), and Levenberg-Marquardt can be used to improve convergence speed and prediction accuracy.
[0043] S44: Based on the O-ring performance degradation model at the actual operating temperature obtained in step S2, the lifetime is introduced into the O-ring lifetime prediction model as a physical law in the form of a function, and the root mean square error is used as the loss function. The physical law is integrated into the loss function in an explicit construction method, specifically including: According to the seal performance degradation model, the life N can be expressed as a function of temperature T, groove depth h1, and compression set retention rate (1-ε): .
[0044] Loss function design: Using RMSE as the loss function, physical laws are explicitly integrated into the loss function: .
[0045] The first term in the above formula is the RMSE calculated for both training and prediction data, and the second term is the RMSE calculated for both physical law prediction (i.e., explicit lifetime formula) and prediction data.
[0046] in, is the compression set retention rate of the O-ring; T is the acceleration temperature; h1 is the groove depth; is the physical weight, used to adjust the degree to which the model conforms to physical laws; N is the lifetime of the O-ring. This represents the true lifetime value of the i-th sample. This represents the predicted lifetime value output by the neural network for the i-th sample; n represents the number of samples in the dataset. For physical degradation models; Model training: By adjusting hyperparameters such as network structure, learning rate, and physical weights, a high-precision lifetime prediction model that balances data fitting and physical constraints is obtained through iterative training.
[0047] S5: Based on the trained O-ring life prediction model, the groove depth of the sealing structure is optimized. The optimization target is the aging life of the O-ring, which can obtain the optimal groove depth parameters and support the selection of the sealing structure.
[0048] This invention introduces the PINN model, which, by embedding physical equations into the loss function of a neural network, can not only fit observed data but also effectively model based on physical laws even with limited sample data, significantly improving the generalization ability and reliability of predictions. The lifespan of sealing rings is affected by multiple factors, among which temperature and groove depth have a crucial impact on material performance degradation and changes in contact state. A high-precision lifespan prediction framework is established based on the PINN model and a performance degradation mechanism model, enabling the screening and evaluation of groove depth parameters at different operating temperatures. This provides strong support for the design and selection of sealing structures, reduces reliance on testing, and improves design efficiency and reliability.
[0049] This invention introduces an O-ring lifetime prediction model based on Physically Informed Neural Networks (PINN). This model combines the powerful data fitting capabilities of neural networks with prior knowledge of physical laws. By incorporating the Arrhenius degradation mechanism as a physical constraint into the PINN framework, the model possesses both physical interpretability and data generalization ability, achieving a deep integration of degradation mechanism and data-driven approaches. Simultaneously, a physical consistency expansion method for multi-condition degradation data is constructed. Under finite-life test sample conditions, the basic dataset is expanded using degradation dynamics, effectively improving the sample diversity for model training and significantly reducing the dependence of lifetime prediction on large amounts of experimental data.
[0050] 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 predicting the lifespan of an aviation O-ring seal, characterized in that: It includes the following steps: S1: Conduct accelerated aging tests: Conduct O-ring life tests, measure the dimensional parameters of O-rings at different aging times and calculate their compression set, obtain O-ring test data under various accelerated temperatures and groove depths, and construct a basic degradation sample set; S2: Constructing an O-ring performance degradation model: Based on the dynamic curve model, construct an O-ring performance degradation model at various acceleration temperatures. Using data-driven regression analysis, calibrate the parameters of the O-ring performance degradation model at various acceleration temperatures according to the basic degradation sample set constructed in step S1, and obtain the O-ring performance degradation model at the actual operating temperature. S3: Life test data expansion: Based on the performance degradation models at various accelerated temperatures constructed in step S2, the basic degradation sample set constructed in step S1 is expanded using a data-driven method to obtain an effective supplementary life test dataset under limited test conditions, thus obtaining a database for machine learning. S4: Construct and train an O-ring lifetime prediction model based on the PINN model, including the following sub-steps: S41: Construct an O-ring lifetime prediction model based on the PINN model. The neural network takes acceleration temperature, groove depth and compression permanent deformation retention rate as input parameters and lifetime as output. Set the number of layers and the number of neurons in each layer of the hidden layer structure, and adjust the parameters of the hidden layer to optimize the fitting ability of the nonlinear relationship between the input parameters and the output. S42: The data in the database used in the machine learning in step S3 enters the neural network from the input layer, is calculated layer by layer through each hidden layer and passed to the output layer to form the prediction result; S43: Calculate the loss function using the backpropagation algorithm combined with gradient descent, and update the weights and biases; S44: Based on the O-ring performance degradation model at actual operating temperature obtained in step S2, the lifetime time is introduced into the O-ring lifetime prediction model as a physical law in the form of a function, and the root mean square error is used as the loss function. The physical law is integrated into the loss function in an explicit construction method: ； in, is the compression set retention rate of the O-ring; T is the acceleration temperature; h1 is the groove depth; is the physical weight, used to adjust the degree to which the model conforms to physical laws; N is the lifetime of the O-ring. This represents the true lifetime value of the i-th sample. This represents the predicted lifetime value output by the neural network for the i-th sample; n represents the number of samples in the dataset. For physical degradation models; By adjusting various hyperparameters, an iterative training method is used to obtain a lifetime prediction model that integrates data fitting and physical constraints.
2. The life prediction method for aviation O-ring seals as described in claim 1, characterized in that: In step S1, the method for calculating the compressive permanent deformation rate of the O-ring is as follows: ； In the formula, h0 is the compression permanent deformation rate of the O-ring; h1 is the original cross-sectional diameter of the O-ring; h2 is the groove depth of the O-ring; h2 is the radial cross-sectional height of the O-ring after deformation.
3. The life prediction method for aviation O-ring seals as described in claim 1, characterized in that: In step S2, the O-ring performance degradation model at actual operating temperature is as follows: ； In the formula, B and α are experimental constants; K is the aging rate constant; and t is the aging time.
4. The life prediction method for aviation O-ring seals as described in claim 3, characterized in that: The method for expanding the basic degradation sample set in step S3 is as follows: the parameters of the O-ring performance degradation model at actual operating temperature are calibrated using data-driven regression analysis and the basic degradation sample set. Using the calibrated O-ring performance degradation model, the existing experimental data are mapped to new temperature and groove depth conditions to generate equivalent degradation curves under untested conditions.
5. The life prediction method for aviation O-ring seals as described in claim 1, characterized in that: The neural network constructed in step S41 calculates weighted inputs: ； In the formula, x is the input parameter; w is the weight; and b is the bias. The output parameters are obtained after being transformed by a transfer function.
6. The life prediction method for aviation O-ring seals as described in claim 1, characterized in that: Step S43 specifically includes the following sub-steps: S431: Error Calculation: Calculate the errors of the output layer and the hidden layer respectively; S432: Weight Update: Update the weights and biases of the neural network based on the gradient of the error function; S433: Method Optimization: Optimize the constructed neural network using one or more of the SGD, Adam, and Levenberg-Marquardt algorithms.
7. The life prediction method for aviation O-ring seals as described in claim 6, characterized in that: The error of the output layer in step S431 is: ； In the formula, L is the loss function constructed in step S44. For the first Layer error term, In a neural network, the first... The linear combination of layers is the input before the activation function is applied.
8. The life prediction method for aviation O-ring seals as described in claim 6, characterized in that: The formula for the hidden layer error propagation in step S431 is: ； in, This represents element-wise multiplication; The derivative of the transfer function; error It is used as the gradient of the loss function for the next neuron in the backpropagation direction, and is passed sequentially until it reaches the input layer.
9. A method for optimizing the structure of an aviation O-ring seal, characterized in that: Based on the aviation O-ring life prediction method according to any one of claims 1-8, the optimal O-ring groove depth parameter is obtained by using the aging life of the O-ring as the optimization target according to the constructed O-ring life prediction model, which is used to support the selection of sealing structure.