Seal ring life prediction method based on multi-physical field coupling

Abstract

This invention relates to the field of digital twin technology, specifically to a method for predicting the lifespan of a sealing ring based on multi-physics coupling. The method includes: acquiring structural loads and environmental parameters of the sealing ring and constructing a digital twin model; extracting deformation state features of the mechanical field and medium diffusion features of the chemical field using encoding operators, and converting them into multi-field heterogeneous vectors; matching stress intensity and diffusion distribution through cross-domain attention units to generate coupling factors between physical fields, dynamically updating the model's permeability parameters, and performing stiffness penalty weighting to generate coupling evolution features; using these coupling evolution features to drive virtual damage accumulation iteration, real-time correction of constitutive parameters until the failure criterion is triggered, and finally outputting lifespan prediction data. This invention solves the problems of isolated simulation fields and large deformation data distortion in sealing rings by constructing a real-time force-chemical field feedback link through cross-domain attention and graph convolution operators.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to a method for predicting the lifespan of sealing rings based on multi-physics coupling. Background Technology

[0002] Currently, using digital twin technology to perform virtual simulation and condition assessment of the sealing rings in precision equipment throughout their entire life cycle has become a research hotspot in the field of modern computer-aided engineering. By constructing a digital twin model that is highly synchronized with the physical entity in virtual space, it is possible to achieve real-time monitoring of the service status of the sealing rings and accurate prediction of their remaining life. However, existing simulation-driven methods for predicting the lifespan of sealing rings still have significant technical shortcomings when constructing high-fidelity digital twin evolution models. In existing numerical simulation frameworks, multi-physics co-simulation typically employs a loosely coupled or step-by-step serial asymmetric architecture. At the program execution level, a single solver is often called first to calculate the initial field distribution of the sealing ring, and then this is used as a static boundary condition input to subsequent solvers. This linear architecture ignores the real-time bidirectional data feedback mechanism between different simulation fields during the long-term service of the sealing ring, causing the digital twin to be in an information island state within the computation step, resulting in the evolution path of the sealing ring in virtual space gradually deviating from the real logic of the physical entity. Meanwhile, in the full life-cycle simulation of the sealing ring, to balance computational efficiency, different simulation domains (such as the deformation domain and the medium diffusion domain) often employ incompatible mesh topologies or asynchronous time steps. Existing technologies, when performing cross-field data transfer, mostly use simple linear interpolation or coarse spatial nearest neighbor mapping algorithms. Due to the strongly nonlinear digital representation properties of the sealing ring, this cross-dimensional heterogeneous data mapping is highly susceptible to inducing energy non-conservation or numerical oscillations in discrete elements. Ultimately, this leads to poor convergence in the simulation of the sealing ring's degradation trajectory, and the prediction accuracy cannot meet the evaluation requirements of high-precision engineering for digital twin systems.

[0003] To address this, a method for predicting the lifespan of sealing rings based on multi-physics coupling is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the life of sealing rings based on multi-physics coupling, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting the life of a sealing ring based on multiphysics coupling includes: The physical geometric load and environmental parameters of the sealing ring are collected, and a digital twin model corresponding to the physical entity is constructed in virtual space. The evolution state of the digital twin model under mechanical and chemical fields is mapped using encoding operators, and deformation state features and medium diffusion features are extracted respectively. At the same time, it is transformed into a multi-field heterogeneous vector through a shared projection matrix. The multi-field heterogeneous feature vectors are input into the cross-domain attention unit, and the deformation stress intensity is captured by the query features. The vectors are then matched with the medium diffusion distribution in the key features to generate coupling factors between physical fields. The permeability parameters of the medium diffusion features in the digital twin model are dynamically updated using the coupling factors, and the deformation evolution features are simultaneously weighted with stiffness penalties using the medium diffusion features to generate coupling evolution features. The coupled evolution characteristics are used to drive the digital twin model to perform virtual damage accumulation iteration, correct the constitutive degradation parameters of the digital twin model in real time and monitor the interface contact pressure distribution; until the geometric seal failure criterion is triggered, the seal ring life prediction data is calculated and output.

[0006] Preferably, the geometric loads include the cross-sectional geometry of the sealing ring, compressibility, filling rate of the sealing groove, fluid pressure, contact preload, and frictional shear force; the environmental parameters include ambient temperature, concentration of chemical components in the medium, pH of the medium, working cycle frequency, and material swelling degree; the digital twin model is a mapping model integrating finite element mesh topology, hyperelastic material constitutive model, and medium diffusion dynamics equations, which is initialized with constraints by the physical geometric loads and environmental parameters, and the force-chemical field state response of the sealing ring is synchronized in real time.

[0007] Preferably, the finite element mesh topology consists of spatial nodes distributed according to the geometric features of the sealing ring and element edges connecting the spatial nodes; The spatial region formed by the spatial nodes and the unit edges is defined as a grid unit; each spatial node serves as a discrete mapping carrier of the physical information of the mechanical and chemical fields in the digital twin model, and is used to store and update the stress tensor set, displacement vector set, medium concentration gradient set, and swelling rate set in real time.

[0008] Preferably, the encoding operator is a feature extraction network based on a graph convolutional neural network, and the extraction process is as follows: The stress tensor set and displacement vector set of the finite element mesh topology in the digital twin model are used as mechanical field inputs. Spatial downsampling and nonlinear feature mapping are performed through the encoding operator to extract deformation state features that characterize the global deformation trend and local stress concentration. Simultaneously, the medium concentration gradient set and swelling rate set of the finite element mesh topology are used as chemical field inputs. Multi-scale feature fusion is performed through the encoding operator to extract medium diffusion features.

[0009] Preferably, the cross-domain attention unit execution process involves linearly projecting the features representing the mechanical field in the multi-field heterogeneous vectors into query feature vectors to capture the deformation stress intensity of a specific grid region in the digital twin model; simultaneously, linearly projecting the features representing the chemical field into bond-value feature vectors and numerical feature vectors to characterize the concentration gradient and permeation potential of the medium diffusion distribution. By calculating the correlation weight matrix between the query feature vector and the key feature vector, the dynamic contribution rate of deformation stress to the permeation behavior of the medium is determined. The correlation weight matrix and the numerical feature vector are then weighted and fused to generate the coupling factor between the physical fields that characterizes the nonlinear interaction intensity between the mechanical field and the chemical field.

[0010] Preferably, the specific process of generating coupling evolution features is to convert the coupling factor into a real-time correction variable of the medium permeability, to repair the diffusion features in the digital twin model, to characterize the accelerating effect of micropore expansion caused by mechanical stress on medium diffusion, and to generate controlled diffusion features. Based on the medium concentration distribution reflected by the controlled diffusion characteristics, the degree of chemical damage to the material at the current moment is evaluated, and the stiffness softening treatment is applied to the deformation evolution characteristics accordingly to simulate the decrease in material load-bearing capacity caused by chemical swelling and generate controlled deformation characteristics. The controlled diffusion features and controlled deformation features are mapped and integrated under the same grid dimension. By extracting nonlinear damage information in the overlapping region, a comprehensive damage index is constructed. The comprehensive damage index is then associated and recombined with the original physical field boundary conditions to generate coupled evolution features.

[0011] Preferably, the specific process of performing virtual damage accumulation iteration is to extract the stress distribution entropy of the sealing interface from the coupling evolution characteristics, quantify the disorder of the contact pressure distribution, and simultaneously calculate the real-time changes of the energy dissipation gradient during mechanical deformation and medium penetration. Based on the time-domain gain coefficient of the fluctuation intensity dynamic mapping of the stress distribution entropy and energy dissipation gradient, the step size of the virtual damage accumulation iteration is nonlinearly reduced or expanded through the time-domain gain coefficient to form a non-equally spaced weighted step sequence. At the end of each iteration step, the deviation between the current evolution trajectory and the preset material decay physical envelope is compared, and step backtracking and real-time correction of the initial gain parameters are triggered based on the deviation magnitude.

[0012] Preferably, the specific process of calculating and outputting the predicted lifespan of the sealing ring is as follows: The total duration of virtual damage evolution is obtained by accumulating the step sequence values ​​before triggering the geometric seal failure criterion during the virtual damage accumulation iteration process. The total duration of the virtual damage evolution is mapped and converted into the number of reciprocating actions aligned with the operating conditions of the physical entity by utilizing the working cycle frequency in the environmental parameters. The coupling evolution characteristics at each moment in the virtual damage accumulation iteration are synchronously correlated to generate and output remaining life prediction data characterizing the performance degradation trajectory of the sealing ring.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention captures the nonlinear correlation between deformation stress and media penetration in real time through cross-domain attention units, breaking the information island state of isolated physical fields in traditional simulations. In the service scenario of the sealing ring, when mechanical compression causes the micropores to open, the coupling factor can dynamically accelerate the penetration logic of the medium and simultaneously provide feedback on the softening effect of chemical swelling on the material's load-bearing capacity. This mutual feedback mechanism enables the digital twin in the virtual space to reproduce the continuous evolution process of pressure-penetration-degradation-failure in real time, ensuring that the simulation path is always anchored within the real physical degradation logic.

[0014] 2. This invention resolves the underlying technical contradiction of energy transfer distortion between heterogeneous meshes during large deformation of the sealing ring by constructing a multi-field feature mapping operator based on topological consistency using a graph convolutional neural network. Unlike traditional algorithms that rely solely on linear interpolation of spatial geometric distances, this invention utilizes the graph convolutional operator to deeply extract the node connectivity relationships and element topological constraints in the finite element mesh of the sealing ring, mapping discrete mechanical response and chemical diffusion data to a continuous vector space with physical structural information. This approach ensures that when the sealing ring undergoes strong nonlinear large displacement, cross-field data exchange always follows the topological correlation laws of the physical structure, eliminating pseudo-numerical fluctuations and false stress concentrations caused by mesh misalignment from a fundamental physical mechanism, and maintaining a strict closed loop of physical quantity conservation relationships during the digital twin evolution process.

[0015] 3. This invention introduces stress distribution entropy and energy dissipation gradient as "sensors" for iterative step size adjustment, achieving sensitive monitoring of the failure risk of the sealing interface. During the severe damage stage near the end of the sealing ring's lifespan, by nonlinearly reducing the step size, it can precisely characterize the transient abrupt behavior of contact pressure loss and medium breaching the sealing barrier. Combined with the real-time backtracking and correction mechanism of the physical envelope, it fundamentally curbs the cumulative deviation of the virtual evolution trajectory over time, ensuring that the output remaining life data perfectly matches the service conditions and performance degradation characteristics of the physical entity. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the method for predicting the life of a sealing ring based on multi-physics coupling. Figure 2This is a schematic diagram of the multi-field heterogeneous feature vector generation process of the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1: Please see Figure 1 This invention provides a method for predicting the life of sealing rings based on multi-physics coupling, the technical solution of which is as follows: A method for predicting the life of a sealing ring based on multiphysics coupling includes: The physical geometric load and environmental parameters of the sealing ring are collected, and a digital twin model corresponding to the physical entity is constructed in virtual space. The evolution state of the digital twin model under mechanical and chemical fields is mapped using encoding operators, and deformation state features and medium diffusion features are extracted respectively. At the same time, it is transformed into a multi-field heterogeneous vector through a shared projection matrix. The multi-field heterogeneous feature vectors are input into the cross-domain attention unit, and the deformation stress intensity is captured by the query features. The vectors are then matched with the medium diffusion distribution in the key features to generate coupling factors between physical fields. The permeability parameters of the medium diffusion features in the digital twin model are dynamically updated using the coupling factors, and the deformation evolution features are simultaneously weighted with stiffness penalties using the medium diffusion features to generate coupling evolution features. The coupled evolution characteristics are used to drive the digital twin model to perform virtual damage accumulation iteration, correct the constitutive degradation parameters of the digital twin model in real time and monitor the interface contact pressure distribution; until the geometric seal failure criterion is triggered, the seal ring life prediction data is calculated and output.

[0019] Collect the physical geometric load and environmental parameters of the sealing ring, and construct a digital twin model corresponding to the physical entity in virtual space; The geometric loads include the cross-sectional geometry of the sealing ring, compressibility, filling rate of the sealing groove, fluid pressure, contact preload, and frictional shear force; the environmental parameters include ambient temperature, concentration of chemical components in the medium, pH of the medium, working cycle frequency, and material swelling degree; the digital twin model is a mapping model integrating finite element mesh topology, hyperelastic material constitutive model, and medium diffusion dynamics equations, which is initialized with constraints by the physical geometric loads and environmental parameters, and the force-chemical field state response of the sealing ring is synchronized in real time.

[0020] Specifically, a 3D scanner is used to measure the cross-sectional diameter and wire diameter of the sealing ring to obtain the basic cross-sectional geometry. Based on the depth and width of the sealing groove, combined with the deformation of the sealing ring after installation, the radial or axial compression ratio is calculated. The filling rate of the sealing groove is determined by the ratio of the sealing ring volume to the cavity volume of the sealing groove. Under actual working conditions, a pressure sensor placed at the end of the sealing cavity collects fluid pressure data in real time, a force gauge is used to obtain the contact preload, and a friction shear force during dynamic operation is obtained through a friction force testing device. Thermocouple temperature sensors are installed in a sealed environment to obtain the ambient temperature in real time. Electrochemical sensors or sampling analyzers are used to monitor the concentration of chemical components and the acidity / alkalinity of the sealing medium. The frequency of equipment operation cycles or rotations is recorded as the working cycle frequency parameter. For specific sealing materials and medium combinations, the mass gain rate of the material at different time points is measured through immersion tests, thereby establishing the evolution relationship between the material swelling degree and the medium concentration over time. In the finite element analysis software environment, a three-dimensional solid model of the sealing ring is established based on the obtained cross-sectional geometry, and meshing is performed to generate a finite element mesh topology structure composed of hexahedral elements. In the material property definition, the constitutive model parameters of the hyperelastic material of the sealing rubber are input. In this embodiment, the parameters of the Moni-Rivlin model are used to describe its nonlinear mechanical characteristics under large deformation. At the same time, the medium diffusion kinetic equation based on Fick's law is embedded in the model to simulate the permeation process of chemical media between rubber molecular chains. The obtained compressibility ratio is converted into displacement boundary conditions and applied to the mesh model. Fluid pressure is applied as a surface load to the water-facing surface of the sealing ring. Ambient temperature and medium chemical component concentration are injected into the model as initial scalar fields to complete constraint initialization. The measured mechanical loads and environmental parameters are input into the model in real time through a data transmission protocol, driving the digital twin model to synchronously update the internal stress distribution, deformation state, and medium diffusion concentration gradient of the sealing ring in each calculation step, thus performing dynamic mapping of the physical force-chemical field state response. By introducing specific geometric loads and multidimensional environmental parameters, this invention achieves a deep mapping of the service state of the sealing ring from the macroscopic mechanical boundary to the microscopic chemical environment. The digital twin model integrates the hyperelastic constitutive model and the medium diffusion dynamics equation, which can accurately capture the nonlinear mechanical characteristics of rubber materials during large deformation processes and simultaneously simulate the material swelling and modification caused by chemical medium penetration.

[0021] The finite element mesh topology consists of spatial nodes distributed according to the geometric features of the sealing ring and element edges connecting the spatial nodes; The spatial region formed by the spatial nodes and the unit edges is defined as a grid unit; each spatial node serves as a discrete mapping carrier of the physical information of the mechanical and chemical fields in the digital twin model, and is used to store and update the stress tensor set, displacement vector set, medium concentration gradient set, and swelling rate set in real time.

[0022] Specifically, based on the pre-obtained cross-sectional geometry, dense spatial nodes are deployed in the virtual space, and adjacent spatial nodes are connected according to a specific topological logic using unit edge lines. In this embodiment, by controlling the length and direction of the control unit edge lines, the sealing ring entity is divided into tens of thousands of hexahedral grid units, and the vertex of each grid unit is a spatial node storing physical information; Each spatial node is assigned a unique data index number, which is defined as a discrete mapping carrier of the physical information of the mechanical and chemical fields in the digital twin model. In the underlying software algorithm, a specific memory space is allocated for each spatial node to store and update the stress tensor set and displacement vector set reflecting the mechanical state in real time. The stress tensor set covers the normal stress components and shear stress components in three spatial directions at the node's coordinates, totaling six independent numerical components; the displacement vector set records the displacement changes of the node relative to its initial position in three spatial dimensions. Simultaneously, a chemical field information storage mechanism is established on the same set of space node carriers. For the medium diffusion process, the medium concentration distribution at each space node is calculated, and the medium concentration gradient set at that node is calculated by using the concentration difference between adjacent nodes and the spatial distance; at the same time, based on the mass increase rate of the material at the current concentration, the corresponding swelling rate set is calculated and stored for each space node. In each time step of the digital twin model calculation, the solver calculates new physical values ​​for each spatial node based on the evolution of boundary conditions. Through a pre-defined data interface, the latest calculated stress tensor, displacement, concentration gradient, and swelling rate are written in real time to the storage register of the corresponding node number. Through this high-frequency discrete mapping and data overlay, the digital twin model can synchronously reconstruct the mechanochemical field state response of the sealing ring during service at the micro-node level. By establishing a discretized mesh topology based on spatial nodes and element edges, a digital physical carrier is provided for the deep integration of mechanical and chemical fields. By storing and dynamically updating multidimensional data such as stress tensor, displacement vector, concentration gradient, and swelling rate in real time at each spatial node, precise tracking of the sealing ring's service status from macroscopic evolution to microscopic local damage is achieved. This discrete mapping mechanism greatly enhances the digital twin model's ability to capture the interaction between local stress concentration and medium penetration, providing data support for the calculation of multiphysics coupling factors.

[0023] See Figure 2 The digital twin model is mapped to its evolution under mechanical and chemical fields using an encoding operator, and deformation state features and medium diffusion features are extracted respectively. The encoding operator is a feature extraction network based on a graph convolutional neural network, and the extraction process is as follows: The stress tensor set and displacement vector set of the finite element mesh topology in the digital twin model are used as mechanical field inputs. Spatial downsampling and nonlinear feature mapping are performed through the encoding operator to extract deformation state features that characterize the global deformation trend and local stress concentration. Simultaneously, the medium concentration gradient set and swelling rate set of the finite element mesh topology are used as chemical field inputs. Multi-scale feature fusion is performed through the encoding operator to extract medium diffusion features.

[0024] Specifically, the finite element mesh topology is transformed into graph structure data. In the graph structure, each spatial node is defined as a vertex of the graph, and the element edges connecting the nodes are defined as edges of the graph. The stress tensor set (containing stress values ​​in 6 directions) and displacement vector set (containing displacement values ​​in 3 dimensions) of each spatial node are concatenated to form the initial attribute matrix of the mechanical field. Simultaneously, the medium concentration gradient set and swelling rate set of each node are concatenated to form the initial attribute matrix of the chemical field, which serves as the dual-channel input of the encoding operator. In the mechanical field feature extraction channel, the encoding operator performs neighborhood information aggregation on the mechanical field attribute matrix to be processed through multi-layer graph convolutional layers. Each convolutional operation weights and fuses the stress and displacement information of adjacent nodes according to the connection relationship of the mesh topology. Subsequently, using spatial downsampling technology (in this embodiment, pooling operation based on node clustering is used), the large-scale mesh node set is compressed into feature clusters with hierarchical structure, thereby capturing the global deformation trend of the sealing ring as a whole. In this process, the data is activated by a nonlinear mapping function, enabling the network to automatically identify and strengthen node regions with extremely large stress tensor values, and extract deformation state features that characterize local stress concentration. The node clustering-based pooling operation divides the original high-density grid into several local node clusters using a clustering algorithm (Graclus algorithm in this embodiment). Within each cluster, the most representative stress and displacement features are extracted using max or average pooling operators. Then, each node cluster is abstracted into a hierarchical supernode to reconstruct a coarsened topology. This effectively preserves key mechanical details such as local stress concentration while achieving multi-scale feature mapping and computational dimensionality reduction from local deformation to overall trend.

[0025] In the chemical field feature extraction channel, the encoding operator uses graph convolution operators with different sensing radii to process the chemical field attribute matrix in parallel. By setting different convolution depths and neighborhood search ranges, molecular permeation patterns at the microscopic level and material swelling evolution data at the macroscopic level are obtained respectively. Specifically, a multi-scale feature fusion mechanism is used to configure parallel sensing radii of 1-2 layers of neighborhood (microscopic), 3-5 layers of neighborhood (mesoscopic), and 8-12 layers of neighborhood (macroscopic), corresponding to convolution depths of 2 to 8 layers and feature channels of 64 to 256 dimensions. A 1×1 convolution operator is used to compress and reconstruct the high-dimensional vectors after splicing each branch. Softmax and layer normalization are introduced to splice and weight the features output by convolutional layers of different depths, eliminating computational noise that may exist at a single scale, and ensuring that the extracted medium diffusion features can completely cover the complete gradient distribution information of the medium permeating from the surface to the core inside the sealing ring. The extracted deformation state features and medium diffusion features are dimensionally aligned through a fully connected layer. The encoding operator, based on preset weight parameters, transforms the nonlinear graph topology features into standardized feature vectors. These feature vectors contain the spatial coupling information of mechanical damage and chemical erosion of the sealing ring at the current moment. The encoding operator uses the finite element mesh topology of the digital twin model as the graph structure basis. It takes the mechanical field attribute matrix formed by splicing the stress tensor set and displacement vector set at each spatial node, and the chemical field attribute matrix formed by splicing the medium concentration gradient set and swelling rate set as inputs, respectively. Through multi-layer graph convolution and spatial downsampling operations, it outputs deformation state feature vectors that characterize the global deformation trend and local stress concentration. Through multi-scale graph convolution and feature fusion, it outputs medium diffusion feature vectors that characterize the medium penetration path and concentration gradient changes. Employing graph convolutional neural networks as encoding operators, this method can directly process irregular finite element mesh topology data, avoiding information loss during data conversion in traditional feature extraction methods. Through spatial downsampling and nonlinear mapping, this method can automatically identify and enhance local stress concentration regions that significantly affect failure from massive amounts of node data; simultaneously, a multi-scale feature fusion mechanism ensures the continuity of information in medium diffusion at both the microscopic penetration and macroscopic swelling levels. Simultaneously, the data is transformed into multi-field heterogeneous vectors through a shared projection matrix, which consists of a set of learnable weight parameters. In this embodiment, the output dimension of the matrix is ​​set to 512 dimensions, and the shared projection matrix is ​​initialized with parameters using a Gaussian distribution. The deformation state feature vector is multiplied with the shared projection matrix, mapping the original data in the mechanical feature space to a unified latent space, generating heterogeneous feature vectors in the mechanical domain. Simultaneously, the medium diffusion feature vector is multiplied with the same shared projection matrix, generating heterogeneous feature vectors in the chemical domain. Because the features of the two physical fields share the same projection operator, the originally independent and physically different heterogeneous data are transformed into the same vector space with the same dimension and scale. The two sets of heterogeneous feature vectors generated are standardized, and the projected vectors are scaled using a normalization operator so that the distribution of feature values ​​conforms to a standard distribution with a mean of 0 and a variance of 1. This eliminates the influence of differences in physical dimensions on subsequent feature matching, and makes the deformation vector of the mechanical field and the diffusion vector of the chemical field completely aligned in mathematical representation.

[0026] The multi-field heterogeneous feature vectors are input into the cross-domain attention unit. The deformation stress intensity is captured using query features and matched with the medium diffusion distribution in the bond value features to generate coupling factors between physical fields. The cross-domain attention unit executes the following process: the feature representing the mechanical field in the multi-field heterogeneous vectors is linearly projected into a query feature vector to capture the deformation stress intensity of a specific grid region in the digital twin model; simultaneously, the feature representing the chemical field is linearly projected into a bond value feature vector and a numerical feature vector to characterize the concentration gradient and osmotic potential of the medium diffusion distribution. By calculating the correlation weight matrix between the query feature vector and the key feature vector, the dynamic contribution rate of deformation stress to the permeation behavior of the medium is determined. The correlation weight matrix and the numerical feature vector are then weighted and fused to generate the coupling factor between the physical fields that characterizes the nonlinear interaction intensity between the mechanical field and the chemical field.

[0027] Specifically, three independent linear transformation weight matrices are preset, defined as the query projection matrix, key-value projection matrix, and numerical projection matrix, respectively. The multi-field heterogeneous feature vector is multiplied with the query projection matrix to generate a 512-dimensional query feature vector. This query feature vector, through the learned weight parameters, focuses on capturing the deformation stress intensity and its spatial distribution in different grid regions of the digital twin model under the current load. Simultaneously, the heterogeneous eigenvectors representing the chemical field are linearly mapped to the bond-value projection matrix and the numerical projection matrix, respectively. Specifically, the heterogeneous chemical field vectors are multiplied by the two sets of weight matrices to generate bond-value eigenvectors and numerical eigenvectors of the same dimension. The bond-value eigenvectors characterize the spatial features of the concentration gradient distribution of the medium within the sealed ring, while the numerical eigenvectors record the quantified values ​​of the medium's permeation potential energy at each grid node at the current moment. The query feature vector and the key feature vector are multiplied by a dot product to calculate the cosine similarity of the two sets of features in the latent space. To prevent gradient explosion or vanishing during the calculation, the dot product result is scaled by dividing it by the square root of the dimension (e.g., the square root of 512, which is approximately 22.6). The scaled value is then processed using an exponential normalization function to generate an association weight matrix with values ​​ranging from 0 to 1. Each element in the association weight matrix represents the dynamic contribution rate of deformation stress at a specific spatial location to the medium's permeation behavior. The correlation weight matrix is ​​multiplied with the numerical feature vector generated above. The stress weight of the mechanical field is used to reconstruct the permeation potential energy of the chemical field in space. The chemical diffusion information corresponding to the grid nodes under stress concentration is enhanced or suppressed. After weighted fusion, a multidimensional vector characterizing the nonlinear interaction strength between the mechanical field and the chemical field is output, namely the coupling factor between the physical fields. By introducing cross-domain attention units, this invention achieves deep adaptive fusion of mechanical strain and chemical diffusion at the feature level; by utilizing the matching operation of query vector and key value vector, it can dynamically identify and quantify the induced enhancement effect of stress concentration region on medium permeation behavior, breaking the computational deadlock of independent information between the two fields in traditional methods.

[0028] The permeability parameter of the medium diffusion feature in the digital twin model is dynamically updated using the coupling factor, and the deformation evolution feature is simultaneously weighted with stiffness penalty using the medium diffusion feature to generate the coupling evolution feature. The specific process of generating coupling evolution features is to convert the coupling factor into a real-time correction variable of the medium permeability, to repair the diffusion features in the digital twin model, to characterize the accelerating effect of micropore expansion caused by mechanical stress on medium diffusion, and to generate controlled diffusion features. Based on the medium concentration distribution reflected by the controlled diffusion characteristics, the degree of chemical damage to the material at the current moment is evaluated, and the stiffness softening treatment is applied to the deformation evolution characteristics accordingly to simulate the decrease in material load-bearing capacity caused by chemical swelling and generate controlled deformation characteristics. The controlled diffusion features and controlled deformation features are mapped and integrated under the same grid dimension. By extracting nonlinear damage information in the overlapping region, a comprehensive damage index is constructed. The comprehensive damage index is then associated and recombined with the original physical field boundary conditions to generate coupled evolution features.

[0029] Specifically, the coupling factor is transformed into a real-time correction variable for medium permeability through a mapping function. This real-time correction variable reflects the change in the material's microstructure caused by mechanical stress. Specifically, when the sealing ring is under pressure or shear, the magnitude of the coupling factor maps the degree of expansion of the micropores inside the material. The construction and application of the mapping function are performed according to the following steps: obtaining the coupling factor, which is a correlation weight matrix between the medium diffusion characteristics and the deformation stress characteristics; since the correlation weight matrix is ​​a normalized value between 0 and 1, the mapping function introduces a preset material physical sensitivity coefficient to convert this normalized value into a permeability increment at the physical scale. The specific logic is as follows: the coupling factor is used as the independent variable, multiplied by the pore expansion constant of the material under pressure, and then weighted and summed with the initial medium permeability; the pore expansion constant is an inherent property of the material measured experimentally.

[0030] The real-time correction variables are used to compensate for and correct the preset initial medium permeability in the digital twin model, thereby obtaining controlled diffusion characteristics. Specifically, the real-time correction variables are mapped one-to-one with the preset initial material permeability constant according to the finite element mesh number. Using product weighted logic, the correction variables are multiplied with the initial permeability as a proportional coefficient. The corrected dynamic permeability is substituted into the diffusion kinetic equation for iteration, and the controlled diffusion characteristics containing the concentration gradient and permeation vector are output.

[0031] Based on the real-time medium concentration distribution information contained in the controlled diffusion characteristics, the degree of chemical damage to the sealing ring material at the current time step is quantitatively assessed. Specifically, a correlation decay curve between material hardness, elastic modulus, and medium concentration is pre-defined. The corresponding stiffness reduction coefficient is retrieved using the concentration values ​​in the controlled diffusion characteristics; the higher the medium concentration, the smaller the reduction coefficient. Based on the reduction coefficient, the deformation evolution characteristics of each mesh element in the digital twin model are subjected to stiffness softening processing. That is, the stiffness reduction coefficient is used to perform a product operation on the material constitutive matrix in the finite element model, directly weakening the tangential stiffness of the mesh elements. The controlled diffusion features and controlled deformation features are mapped and integrated under the same grid dimension. By extracting nonlinear damage information from overlapping regions, a comprehensive damage index is constructed. The mapping and integration is achieved through a spatial alignment algorithm, using the finite element grid number in the digital twin model as a unique index. The concentration gradient vector in the controlled diffusion features and the stress tensor vector in the controlled deformation features are superimposed in union to construct a high-dimensional heterogeneous damage vector. Through a preset nonlinear discrimination criterion, grid cells overlapping stress concentration areas and high-permeability areas of the medium are identified. The nonlinear discrimination criterion is to extract the equivalent stress and medium concentration of each grid cell in the digital twin model, and normalize them by dividing them by preset material yield strength and saturation concentration to obtain numerical coefficients between 0 and 1. Then, a bidirectional logical threshold is set to determine grid numbers with stress level coefficients greater than or equal to 0.7 and concentration saturation coefficients greater than or equal to 0.5 as highly sensitive cells. By performing a logical AND operation on the entire grid, overlapping grid regions that simultaneously meet the above two threshold conditions are identified and marked as nonlinear coupling damage regions. Using the finite element mesh number in the digital twin model as a unique spatial index, the controlled diffusion feature vector containing the medium concentration distribution and the controlled deformation feature vector containing the stress tensor distribution are dimensionally aligned. Element-level multiplication and weighting operations are performed: for the identified overlapping mesh elements, the concentration gradient value and the equivalent stress value corresponding to that position are multiplied to generate the initial coupled damage amount. A damage amplification factor with a value between 1.2 and 1.5 is introduced as a weighting factor and multiplied with the initial coupled damage amount to quantify the nonlinear damage increment caused by the accelerated medium penetration due to mechanical stress. The calculation results are processed by modulus norm and normalized to between 0 and 1, defined as a comprehensive damage index. The damage amplification factor is calibrated based on material experimental data. Specifically, fatigue tests on the sealing ring material under pure mechanical load and chemical degradation tests under pure medium immersion are conducted beforehand to obtain damage rate curves under two single-field conditions. Accelerated aging tests are then performed under a force-chemical coupling environment. By comparing the deviation between the coupled experimental data and the superimposed data from the previous two methods, the additional damage increment caused by the coupling effect is calculated. The ratio of this increment to the total damage of a single physical field is defined as the basic amplification factor. Based on experimental calibration, corresponding values ​​are selected within the range of 1.2 to 1.5 for different operating frequencies and medium pH levels. For example, when the operating cycle frequency increases or the medium corrosivity increases, the damage amplification factor is set towards the upper limit of 1.5 to realistically simulate the sharp drop in material performance under strong coupling conditions.

[0032] The comprehensive damage index is correlated and recombined with the original physical field boundary conditions to generate coupled evolution characteristics. The correlation and recombination is achieved by aligning the grid numbers and mapping the comprehensive damage index of each unit to the corresponding boundary action surface. A critical damage threshold is preset (set to 0.6 in this embodiment). When the comprehensive damage index is lower than the threshold, the original fluid pressure and preload load remain unchanged. Once the index exceeds the threshold, the load in the region is accelerated by using an exponential function. In practice, the grid number of severely damaged areas is retrieved in real time. Once the damage index at that location exceeds the preset threshold, the fluid pressure load or contact preload value at that location is automatically reduced using the above logic to simulate the decrease in load-bearing capacity or sealing leakage caused by material deterioration. These corrected dynamic boundary parameters are then recombined with the current stress and concentration data and encapsulated into a unified feature vector, which generates coupled evolution features.

[0033] This application constructs a two-way feedback loop at the underlying logic level, which is a mechanical stress-induced micropore expansion and a chemical swelling-induced stiffness softening mechanism. This achieves deep decoupling and dynamic reconstruction of the seal ring damage evolution mechanism. It uses a spatial alignment algorithm and a nonlinear discrimination criterion to accurately locate the high-risk region of force-chemical field superposition. Furthermore, it scientifically characterizes the synergistic degradation effect of the material under the coexistence of multiple fields through a damage amplification factor, effectively compensating for the distortion of physical mechanism caused by traditional single-field analysis or linear superposition.

[0034] The specific process of performing virtual damage accumulation iteration is to extract the stress distribution entropy of the sealing interface from the coupling evolution characteristics, quantify the disorder of the contact pressure distribution, and simultaneously calculate the real-time changes of energy dissipation gradient during mechanical deformation and medium penetration. Based on the time-domain gain coefficient of the fluctuation intensity dynamic mapping of the stress distribution entropy and energy dissipation gradient, the step size of the virtual damage accumulation iteration is nonlinearly reduced or expanded through the time-domain gain coefficient to form a non-equally spaced weighted step sequence. At the end of each iteration step, the deviation between the current evolution trajectory and the preset material decay physical envelope is compared, and step backtracking and real-time correction of the initial gain parameters are triggered based on the deviation.

[0035] Contact stress data at the sealing interface is extracted from the coupled evolution characteristics. The uniformity of the contact pressure distribution is quantified using information entropy theory. By calculating the dispersion and distribution density of stress at each grid node on the sealing surface, stress distribution entropy is generated. The larger the entropy value, the more disordered the contact pressure distribution and the higher the risk of seal failure. Simultaneously, the strain energy change during mechanical deformation and the chemical potential energy loss during medium permeation are monitored in real time, and the rate of change of both with the iteration step is defined as the energy dissipation gradient. A set of initial gain parameters is preset, including an initial permeability gain coefficient (preset to 1.0) and an initial material attenuation gain factor (preset to 0.05); the initial gain parameters determine the advancement speed of the coupling evolution characteristics in the time domain; the initial gain parameters are used to weight the medium permeation velocity and stiffness softening degree at the initial moment as the starting point benchmark for virtual damage iteration; The time-domain gain coefficient is determined based on the changes in stress distribution entropy and energy dissipation gradient. The base step size is set to one hundred hours. When the stress distribution entropy is less than 1.5 and the energy dissipation gradient fluctuation is small, the time domain gain coefficient is set to 1.5, and the virtual time step of a single iteration is expanded to 150 hours; when the stress distribution entropy is between 1.5 and 2.2, the coefficient is set to 0.8, and the step size is reduced to 80 hours; when the stress distribution entropy exceeds 2.2, the coefficient is set to 0.2, and the step size is compressed to 20 hours. The dynamically adjusted step sizes are arranged in chronological order to form a set of non-equal interval weighted step sequences; After each step iteration, the current comprehensive damage index is extracted and compared with the preset material decay physical envelope; the geometric distance from the current data point to the standard curve is calculated in real time and defined as the deviation; if the deviation is greater than 5% of the standard value, it means that the current initial gain parameter can no longer accurately describe the actual decay rate of the material. When the deviation exceeds the standard, the current virtual time is forcibly rolled back to the last iteration node that met the standard. At this time, the initial gain parameters are corrected online according to the positive or negative direction of the deviation: if the damage calculation is faster than the standard envelope, the initial permeability gain coefficient is reduced by 0.03; if the damage evolution is slower than the standard, the initial material attenuation gain factor is increased by 0.08. The corrected parameters will replace the original initial values, and the calculation of this step size will be re-executed until the deviation converges to within 2%.

[0036] By employing a virtual damage accumulation iterative method, dynamic adaptation of the simulation evolution step size to the actual degradation rhythm of the material is achieved. By utilizing stress distribution entropy and energy dissipation gradient to perceive the physical fluctuations of the sealing interface in real time, the iterative process can automatically perform nonlinear scaling of the step size according to the severity of damage evolution. This effectively solves the technical bottlenecks of traditional fixed step sizes, which are prone to divergence during periods of rapid evolution and computational redundancy during periods of stability. Furthermore, a step-back and self-correction mechanism based on the physical envelope ensures that the multi-field coupling trajectory is always constrained within the reasonable range of materials science, enhancing the numerical convergence and physical consistency of the complex nonlinear simulation process.

[0037] The specific process for calculating and outputting the predicted lifespan of the sealing ring is as follows: The total duration of virtual damage evolution is obtained by accumulating the weighted step sequence values ​​during the virtual damage accumulation iteration process. The total duration of the virtual damage evolution is mapped to a time scale using the working cycle frequency in the environmental parameters to generate the number of reciprocating actions aligned with the operation of the physical entity. The coupling evolution characteristics at each moment in the virtual damage accumulation iteration are synchronously correlated to generate and output sealing ring lifetime prediction data characterizing the sealing ring performance degradation trajectory; A non-equal-interval weighted step sequence formed during the virtual damage accumulation iteration process is obtained. This sequence records each evolution step of the digital twin model in the simulation environment. The size of each step is corrected in real time by stress distribution entropy and energy dissipation gradient. By accumulating all the weighted step values ​​in this sequence, a total value can be obtained. This value represents the total virtual damage evolution time experienced by the sealing ring from the initial state to the preset failure judgment boundary (such as the sealing contact stress being lower than the critical threshold) under the multi-field coupling effect. This total time reflects the material's tolerance under the simulated high-dimensional physical field logic. To convert the evolution time in virtual space into a lifespan indicator that can be referenced in actual engineering, the working cycle frequency from pre-collected environmental parameters is retrieved. Based on the motion characteristics of the physical entity under actual working conditions, the total duration of virtual damage evolution is used as a time benchmark, and the working cycle frequency is used to perform time-scale mapping. Specifically, the total duration is multiplied by the frequency for proportional conversion, thereby transforming the abstract virtual time unit into a specific number of reciprocating movements or rotations. While calculating the number of actions, the coupled evolutionary features of each key time node in the virtual damage accumulation iteration are simultaneously extracted and correlated. These features include chemical damage data generated by controlled diffusion features at each stage and stiffness weakening data generated by controlled deformation features. By arranging these multidimensional features in chronological order, a performance evolution curve that changes with the number of reciprocating actions is constructed. This performance evolution curve can clearly show the evolutionary logic of the sealing ring from the elastic stage, the damage initiation stage to the final failure stage. The performance degradation trajectory of the sealing ring is extracted by using the generated performance evolution curve. This trajectory not only includes the value at the end of the life, but also the slope changes of key performance indicators (such as contact pressure distribution, media penetration depth, and material effective stiffness) with time. By analyzing the changing trend in the trajectory, the inflection point of damage acceleration is identified, thereby generating a performance degradation data package characterizing the entire life cycle of the sealing ring. The calculated number of reciprocating motions, performance degradation trajectory, and physical field distribution of key nodes are integrated to generate and output the final seal ring life prediction data.

[0038] By calculating and outputting the sealing ring life data, the simulation evolution results were directly mapped to the engineering application indicators. The abstract virtual evolution duration was transformed into an intuitive number of reciprocating actions by utilizing the working cycle frequency, effectively bridging the perceptual gap between numerical simulation and the physical entity's operating pace.

[0039] This invention establishes a real-time bidirectional feedback mechanism between deformation stress intensity and medium diffusion behavior by constructing a digital twin model and introducing cross-domain attention units. This breaks the traditional architecture of isolated physical fields in simulation, thereby restoring the physical degradation logic under force-chemical interaction at the mechanism level. By using encoding operators and shared projection matrices to map the heterogeneous features of multiple fields to a unified vector space, the invention avoids the risks of energy non-conservation and numerical fluctuations caused by grid topology incompatibility or coarse spatial mapping. This ensures the logical stability and physical self-consistency of complex nonlinear evolution paths in the calculation process, providing a digital characterization scheme with bidirectional feedback capability for the state assessment of sealed systems during long-term service.

[0040] Example 2: This embodiment applies the sealing ring life prediction method based on multi-physics field coupling to the prediction of the sealing ring life of hydraulic support in deep-sea oil and gas drilling platform. This embodiment selects the nitrile rubber sealing ring used in deep-sea hydraulic supports as the object. First, initial parameters such as its cross-sectional diameter of 5.33 mm, hydraulic pressure of 21 MPa, and ambient temperature of 85°C are collected. In virtual space, a finite element mesh model with 52,400 nodes is constructed, and the parameters of the hyperelastic material constitutive model and the medium diffusion dynamics equation are loaded to complete the constraint initialization of the model, ensuring that the model can synchronize the mechanical and chemical state response of the physical entity in real time. Using a graph convolutional neural network-based encoding operator, spatial sampling is performed on the digital twin model; the stress tensor and displacement vector of each node of the sealing ring are extracted through the mechanical channel (identifying the maximum stress at the lip as 24.5 MPa), while the medium concentration gradient and swelling rate are extracted simultaneously through the chemical channel; by sharing the projection matrix, these heterogeneous physical data are transformed into 512-dimensional feature vectors, enabling the comparison of mechanical deformation and chemical permeation characteristics within a unified mathematical space; The aforementioned feature vectors are input into the cross-domain attention unit, where mechanical features are used as query terms and chemical features are used as key terms for matching operations. Through calculation, it is found that in high-stress regions, the micropore expansion caused by mechanical compression will significantly change the permeation path of the medium. Through this matching operation, a coupling factor that quantifies the interaction strength between physical fields is generated, accurately identifying the dynamic contribution rate of stress concentration to the permeation behavior of the medium. By dynamically modifying the medium permeability parameters in the digital twin model using coupling factors, the physical process of stress-accelerated permeation is simulated, generating controlled diffusion characteristics. Simultaneously, based on the real-time medium concentration distribution, stiffness penalty weighting is applied to the material to simulate material softening and load-bearing capacity reduction caused by chemical swelling (stiffness is reduced by 18% in this embodiment), generating controlled deformation characteristics. Finally, the two are mapped and integrated to construct a comprehensive evolutionary characteristic containing nonlinear damage information. The driving model performs virtual damage iterations, extracting the stress distribution entropy and energy dissipation gradient of the sealing interface from the coupling features. When drastic physical evolution is detected (fluctuations exceeding a preset threshold), the iteration step size is automatically reduced nonlinearly from 50 hours to 5 hours to precisely capture the failure transient. After each iteration, the physical envelope of material decay is compared. If the deviation exceeds 15%, step backtracking is triggered and parameters are corrected to ensure that the simulation path conforms to the real material degradation logic. The contact pressure distribution at the sealing interface is monitored in real time. When the effective contact bandwidth decreases to less than 30% of the original width, the values ​​of each step sequence are accumulated to obtain a total virtual damage duration of 3250 hours. Combined with the working cycle frequency (15 reciprocating actions per minute), the total duration is converted into the total number of service cycles of the physical entity (2.925 million cycles), and the remaining life prediction curve and safe service factor are output.

[0041] 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 predicting the lifespan of a sealing ring based on multiphysics coupling, characterized in that, include: The physical geometric load and environmental parameters of the sealing ring are collected, and a digital twin model corresponding to the physical entity is constructed in virtual space. The evolution state of the digital twin model under mechanical and chemical fields is mapped using encoding operators, and deformation state features and medium diffusion features are extracted respectively. At the same time, it is transformed into a multi-field heterogeneous vector through a shared projection matrix. The multi-field heterogeneous feature vectors are input into the cross-domain attention unit, and the deformation stress intensity is captured by the query features. The vectors are then matched with the medium diffusion distribution in the key features to generate coupling factors between physical fields. The permeability parameters of the medium diffusion features in the digital twin model are dynamically updated using the coupling factors, and the deformation evolution features are simultaneously weighted with stiffness penalties using the medium diffusion features to generate coupling evolution features. The coupled evolution characteristics are used to drive the digital twin model to perform virtual damage accumulation iteration, trigger step backtracking and real-time correction of the initial gain parameters, and calculate and output the sealing ring life prediction data.

2. The sealing ring life prediction method based on multiphysics coupling according to claim 1, characterized in that, The geometric loads include the cross-sectional geometry of the sealing ring, compressibility, filling rate of the sealing groove, fluid pressure, contact preload, and frictional shear force; the environmental parameters include ambient temperature, concentration of chemical components in the medium, pH of the medium, working cycle frequency, and material swelling degree; the digital twin model is a mapping model integrating finite element mesh topology, hyperelastic material constitutive model, and medium diffusion dynamics equations, which is initialized with constraints by the physical geometric loads and environmental parameters, and the force-chemical field state response of the sealing ring is synchronized in real time.

3. The sealing ring life prediction method based on multiphysics coupling according to claim 2, characterized in that, The finite element mesh topology consists of spatial nodes distributed according to the geometric features of the sealing ring and element edges connecting the spatial nodes; The spatial region formed by the spatial nodes and the unit edges is defined as a grid unit; each spatial node serves as a discrete mapping carrier of the physical information of the mechanical and chemical fields in the digital twin model, and is used to store and update the stress tensor set, displacement vector set, medium concentration gradient set, and swelling rate set in real time.

4. The sealing ring life prediction method based on multiphysics coupling according to claim 1, characterized in that, The encoding operator is a feature extraction network based on a graph convolutional neural network, and the extraction process is as follows: The stress tensor set and displacement vector set of the finite element mesh topology in the digital twin model are used as mechanical field inputs. Spatial downsampling and nonlinear feature mapping are performed through the coding operator to extract deformation state features that characterize the global deformation trend and local stress concentration. Simultaneously, the medium concentration gradient set and swelling rate set of the finite element mesh topology are used as chemical field inputs, and multi-scale feature fusion is performed through the coding operator to extract medium diffusion features.

5. The sealing ring life prediction method based on multiphysics coupling according to claim 1, characterized in that, The cross-domain attention unit execution process involves linearly projecting the features representing the mechanical field in the multi-field heterogeneous vectors into query feature vectors, which are used to capture the deformation stress intensity of a specific grid region in the digital twin model. Simultaneously, the characteristic linear projections representing the chemical field are transformed into bond-value eigenvectors and numerical eigenvectors, which are used to characterize the concentration gradient and osmotic potential of the medium diffusion distribution. By calculating the correlation weight matrix between the query feature vector and the key feature vector, the dynamic contribution rate of deformation stress to the permeation behavior of the medium is determined. The correlation weight matrix and the numerical feature vector are then weighted and fused to generate the coupling factor between the physical fields that characterizes the nonlinear interaction intensity between the mechanical field and the chemical field.

6. The sealing ring life prediction method based on multiphysics coupling according to claim 1, characterized in that, The specific process of generating coupling evolution features is to convert the coupling factor into a real-time correction variable of the medium permeability, to repair the diffusion features in the digital twin model, to characterize the accelerating effect of micropore expansion caused by mechanical stress on medium diffusion, and to generate controlled diffusion features. Based on the medium concentration distribution reflected by the controlled diffusion characteristics, the degree of chemical damage to the material at the current moment is evaluated, and the stiffness softening treatment is applied to the deformation evolution characteristics accordingly to simulate the decrease in material load-bearing capacity caused by chemical swelling and generate controlled deformation characteristics. The controlled diffusion features and controlled deformation features are mapped and integrated under the same grid dimension. By extracting nonlinear damage information in the overlapping region, a comprehensive damage index is constructed. The comprehensive damage index is then associated and recombined with the original physical field boundary conditions to generate coupled evolution features.

7. The sealing ring life prediction method based on multiphysics coupling according to claim 1, characterized in that, The specific process of performing virtual damage accumulation iteration is to extract the stress distribution entropy of the sealing interface from the coupling evolution characteristics, quantify the disorder of the contact pressure distribution, and simultaneously calculate the real-time changes of energy dissipation gradient during mechanical deformation and medium penetration. Based on the time-domain gain coefficient of the fluctuation intensity dynamic mapping of the stress distribution entropy and energy dissipation gradient, the step size of the virtual damage accumulation iteration is nonlinearly reduced or expanded through the time-domain gain coefficient to form a non-equally spaced weighted step sequence. At the end of each iteration step, the deviation between the current evolution trajectory and the preset material decay physical envelope is compared, and step backtracking and real-time correction of the initial gain parameters are triggered based on the deviation.

8. The sealing ring life prediction method based on multiphysics coupling according to claim 1, characterized in that, The specific process for calculating and outputting the predicted lifespan of the sealing ring is as follows: The total duration of virtual damage evolution is obtained by accumulating the weighted step sequence values ​​during the virtual damage accumulation iteration process. The total duration of the virtual damage evolution is mapped to a time scale using the working cycle frequency in the environmental parameters to generate the number of reciprocating actions aligned with the operation of the physical entity. The coupling evolution characteristics at each moment in the virtual damage accumulation iteration are synchronously correlated to generate and output sealing ring life prediction data characterizing the sealing ring performance degradation trajectory.

Images