Digital twin analysis method for corrosion resistance of suspension damping bushing

Abstract

The present application relates to the technical field of digital twinning, and discloses a digital twinning analysis method for the corrosion resistance of a suspension damping bushing. The method obtains vibration displacement time series data and multi-dimensional corrosion factor sequence data of the suspension damping bushing; the corrosion factor sequence is resampled in the time dimension with a vibration period as a reference, and the corrosion factor data is mapped to the spatial coordinate nodes of a three-dimensional digital twinning grid model of the bushing in the spatial dimension, to generate a spatio-temporal aligned multi-physical field fusion data set; the fusion data set is input into a digital twinning model, the correlation between stress transmission characteristics and corrosion medium diffusion characteristics between grid nodes is extracted by a graph neural network, and corrosion layer degradation state data is output. The present application eliminates the spatio-temporal misalignment error between vibration data and corrosion data, enables the digital twinning model to reflect the physical mechanism of alternating stress accelerated corrosion, overcomes the prediction distortion defects caused by independent channel analysis, and improves the accuracy and reliability of corrosion resistance evolution analysis.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology and discloses a digital twin analysis method for the corrosion resistance performance of suspension damping bushings. Background Technology

[0002] In the field of vehicle component data processing technology, the corrosion performance analysis of suspension damping bushings typically employs digital twin models for simulation. Existing conventional solutions treat structural mechanics analysis and environmental corrosion analysis as two independent data processing channels when constructing the digital twin analysis model of suspension damping bushings. Specifically, this involves: collecting vibration displacement data of the bushing during vehicle operation using sensors, and directly inputting this data into the structural mechanics simulation mesh model to calculate stress distribution; simultaneously, acquiring temperature, humidity, and salt spray concentration data of the bushing's external environment through environmental monitoring equipment, and inputting these environmental parameters as static boundary conditions into an independent corrosion simulation model. After obtaining the respective calculation results, a simple spatial overlay comparison is performed between the stress concentration region obtained from the mechanical simulation and the corrosion rate distribution region obtained from the corrosion simulation to determine the degradation of the anti-corrosion layer. This independent channel analysis method does not synchronize the mechanical data and environmental corrosion data in terms of time and space dimensions during the data input stage; instead, it assumes that the two data interact unidirectionally without interference at their respective time steps and spatial mesh nodes.

[0003] The core technical problem with the aforementioned existing solutions lies in the following: Since alternating stress dynamically accelerates the diffusion of corrosive media, inputting vibration displacement time-series data and environmental corrosion data separately into independent simulation channels prevents the digital twin model from reflecting the physical mechanism of stress-corrosion coupling, leading to distorted predictions of corrosion resistance evolution. In actual service, the stress changes within each vibration cycle of the suspension damping bushing alter the opening and closing state of microcracks in the corrosion protection layer, changing the diffusion path and rate of the corrosive medium within the layer in real time. Existing technologies lack a data flow mechanism for spatiotemporally aligning and coupling mechanical and corrosion characteristics, causing a misalignment between the model-calculated corrosion medium diffusion state and the actual dynamic physical evolution process, resulting in the final output data on corrosion layer degradation deviating from actual physical laws. Summary of the Invention

[0004] The purpose of this invention is to provide a digital twin analysis method for the corrosion resistance of suspension damping bushings, which can solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A digital twin analysis method for the corrosion resistance of suspension damping bushings includes: acquiring the vibration displacement time series data of the suspension damping bushings during service and the multidimensional corrosion factor sequence data of the environment in which they are located; In the time dimension, the multidimensional corrosion factor sequence data is resampled based on the vibration period in the vibration displacement time series data to obtain a time-aligned corrosion factor sequence. In the spatial dimension, the corrosion factor data in the time-aligned corrosion factor sequence is mapped to the corresponding spatial coordinate nodes in the three-dimensional digital twin mesh model of the suspension damping bushing to generate a spatiotemporally aligned multiphysics fusion dataset. The multiphysics fusion dataset is input into a pre-constructed digital twin model. The graph neural network in the digital twin model is used to extract the correlation between stress transfer characteristics and corrosion medium diffusion characteristics between grid nodes in the three-dimensional digital twin mesh model. Based on the correlation, the degradation status data of the anti-corrosion layer of the suspension damping bushing at different spatial locations are output.

[0006] Preferably, the step of resampling the multidimensional corrosion factor sequence data based on the vibration period in the vibration displacement time series data includes: performing a Fourier transform on the vibration displacement time series data to extract the dominant frequency component in the frequency domain signal; The duration of a single vibration cycle is calculated based on the dominant frequency component. Using the duration of a single vibration cycle as a time window, the multidimensional corrosion factor sequence data is divided into multiple corrosion factor subsequences; Calculate the mean of each corrosion factor subsequence within the corresponding time window, and use the mean as the corrosion factor value at the center of the time window to generate the time-aligned corrosion factor sequence.

[0007] Preferably, the step of mapping the corrosion factor data in the time-aligned corrosion factor sequence to the corresponding spatial coordinate nodes in the three-dimensional digital twin mesh model of the suspension damping bushing includes: acquiring the physical spatial coordinates of multiple environmental sensors arranged around the suspension damping bushing and the raw corrosion factor data collected by the environmental sensors. Using the inverse distance weighted interpolation algorithm, with the physical spatial coordinates of the environmental sensor as the interpolation nodes and the original corrosion factor data as the interpolation attribute, the interpolated corrosion factor data at each spatial coordinate node in the three-dimensional digital twin mesh model is calculated. The interpolated corrosion factor data at the same time are assigned to the corresponding spatial coordinate nodes.

[0008] Preferably, the generation of the spatiotemporally aligned multiphysics fusion dataset includes: extracting the stress response features corresponding to each spatial coordinate node in the three-dimensional digital twin mesh model in the vibration displacement time series data; The stress response characteristics and the corrosion factor data mapped to the same spatial coordinate node are subjected to dimensionless processing to obtain a normalized stress feature vector and a normalized corrosion factor vector. The normalized stress feature vector and the normalized corrosion factor vector are concatenated along the feature dimension to obtain the multi-physics node feature vector of the spatial coordinate node. The multiphysics fusion dataset is composed of the feature vectors of all the spatial coordinate nodes.

[0009] Preferably, the step of extracting the correlation between stress transfer characteristics and corrosion medium diffusion characteristics between grid nodes in the three-dimensional digital twin mesh model through the graph neural network in the digital twin model includes: constructing a graph topology structure with spatial coordinate nodes in the three-dimensional digital twin mesh model as graph nodes and physical connection relationships between adjacent spatial coordinate nodes as graph edges; The initial physical weights of the graph edges are calculated based on the spatial distance between adjacent graph nodes and the material permeability parameter. The graph neural network, based on the graph topology and the initial physical weights, extracts the association relationship by passing node features from the multiphysics fusion dataset to the graph edges through graph convolutional layers.

[0010] Preferably, the step of outputting the corrosion degradation status data of the suspension damping bushing at different spatial locations according to the correlation includes: inputting the correlation into the fully connected layer of the digital twin model, and calculating the current thickness attenuation and adhesion strength attenuation of the corrosion layer at each spatial coordinate node; The current thickness attenuation and the adhesion strength attenuation are input into a preset micro-damage mapping function to calculate the microcrack propagation probability at the spatial coordinate node. Based on the microcrack propagation probability and the preset degradation threshold range, the degradation level of the anti-corrosion layer to which each spatial coordinate node belongs is determined, and the degradation level of the anti-corrosion layer of all spatial coordinate nodes is output as the degradation status data of the anti-corrosion layer.

[0011] Preferably, the extraction of the dominant frequency component in the frequency domain signal includes: dividing the vibration displacement time series data into multiple sliding time window data with overlapping regions; Perform a short-time Fourier transform on each of the sliding time window data to obtain a time-spectrum graph; In the time-frequency spectrum, the frequency change trajectory corresponding to the energy peak is traced along the time axis, and the frequency corresponding to the energy peak within the current sliding time window is determined as the dynamic dominant frequency at the current moment. The step of calculating the duration of a single vibration cycle based on the dominant frequency component includes: performing a smoothing filter on the dynamic dominant frequency within adjacent sliding time windows, and calculating the duration of a dynamic single vibration cycle that changes with time based on the reciprocal of the filtered dynamic dominant frequency.

[0012] Preferably, the step of using the inverse distance weighted interpolation algorithm to calculate the interpolated corrosion factor data at each spatial coordinate node in the three-dimensional digital twin mesh model includes: calculating the initial Euclidean distance between the physical spatial coordinates of the environmental sensor and the spatial coordinate node; Obtain the surface curvature data of the three-dimensional digital twin mesh model along the direction of the initial Euclidean distance line; The initial Euclidean distance is corrected using the surface curvature data to obtain the surface correction distance, wherein the greater the surface curvature, the longer the surface correction distance. The surface correction distance is used to replace the initial Euclidean distance as the distance parameter in the inverse distance weighted interpolation algorithm to calculate the interpolated corrosion factor data.

[0013] Preferably, the step of transmitting node features in the multiphysics fusion dataset on the graph edges via graph convolutional layers includes: using an expanded graph convolutional layer with multiple expansion rates to process the node features in the multiphysics fusion dataset in parallel, wherein different expansion rates correspond to different hop counts of adjacent graph node feature aggregation ranges. The local aggregated features output from the convolutional layers of the expanded maps with different expansion rates are stitched together; The spliced ​​local aggregated features are input into the residual network layer, and the residual network layer adds the spliced ​​local aggregated features to the initial node features in the multiphysics fusion dataset to obtain the global association features output by the graph convolutional layer.

[0014] Preferably, the step of inputting the current thickness attenuation and the adhesion strength attenuation into a preset micro-damage mapping function includes: inputting the current thickness attenuation, the adhesion strength attenuation, and the microcrack propagation probability sequence of the spatial coordinate node at a historical moment into a long short-term memory network. The forgetting gating unit of the long short-term memory network discards the historical microcrack propagation probability that is irrelevant to the stress state at the current moment. The cell state is updated by combining the current thickness decay amount and the adhesion strength decay amount at the current moment using the input gating unit of the long short-term memory network; The updated cell state is output through the output gating unit as the output probability of the dynamically adjusted micro-damage mapping function.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention solves the problem of distorted prediction of corrosion resistance performance evolution by introducing a spatiotemporal alignment mechanism into a digital twin model. The scheme uses the vibration period in the vibration displacement time series data as a benchmark, resamples the multidimensional corrosion factor sequence data, and maps the corrosion factor data to the corresponding spatial coordinate nodes of the three-dimensional digital twin mesh model in the spatial dimension, generating a spatiotemporally aligned multiphysics fusion dataset. The construction of this dataset breaks the separation between independent input of mechanical and corrosion data in existing technologies, enabling the data input to the graph neural network to have the same time and spatial coordinate benchmarks. Based on this fusion dataset, the graph neural network extracts the correlation between stress transfer characteristics and corrosion medium diffusion characteristics between mesh nodes, allowing the digital twin model to perform calculations based on the physical mechanism of alternating stress and corrosion medium coupling. The output corrosion layer degradation state data reflects the real dynamic process of stress-accelerated corrosion, eliminating prediction misalignment errors caused by independent channel analysis.

[0016] 2. Based on spatiotemporal alignment, this invention further optimizes the accuracy of calculation results by refining data processing and model structure. By extracting the dominant frequency component from the vibration displacement time-series data using Fourier transform or short-time Fourier transform, it can adapt to the dynamic changes in vibration period under different driving conditions, ensuring that the time alignment benchmark is consistent with the actual physical vibration state. In the spatial mapping stage, an inverse distance weighted interpolation algorithm is introduced and Euclidean distance is corrected by combining surface curvature data, so that the corrosion factor data calculated by interpolation fits the geometric morphology features of the bushing surface. In the feature extraction stage, the features of adjacent nodes with different hop numbers are processed in parallel by a multi-expansion-rate expansion graph convolutional layer, and the initial node features are retained by the residual network layer, avoiding the loss of local features during the transmission process of deep networks. In the degradation state calculation stage, the current state is updated by using a long short-term memory network combined with the microcrack propagation probability at historical moments, so that the microcrack propagation probability output by the model includes the cumulative decay logic of the time dimension, improving the reliability of the analysis of the evolution trend of local hidden damage in the anti-corrosion layer. Attached Figure Description

[0017] Figure 1 Main flowchart of digital twin analysis method for corrosion resistance performance of suspension damping bushings; Figure 2 Generate a flowchart for time-aligned corrosion factor sequences; Figure 3 This is a flowchart of spatial coordinate mapping based on surface correction; Figure 4 Flowchart for generating multiphysics fusion datasets; Figure 5 This is a flowchart of the graph neural network feature extraction and association calculation process; Figure 6 It is a flow chart for outputting data on the degradation state of the anti-corrosion layer. Specific implementation manners

[0018] Please refer to Figure 1 , this embodiment provides a digital twin analysis method for the anti-corrosion performance of a suspension damping bushing. The suspension damping bushing is a damping bushing with a rubber-metal composite structure in a vehicle suspension system. An organic anti-corrosion coating is provided on the outer surface of the metal skeleton of the bushing. The bushing serves during the vehicle driving process,承受来自悬挂系统的交变振动载荷,同时暴露于包含多种腐蚀影响因素的环境介质中。

[0019] Obtain the vibration displacement time series data of the suspension damping bushing during its service life and the multi-dimensional corrosion factor sequence data of the environment where it is located. The vibration displacement time series data is obtained by a displacement acquisition unit arranged between the metal inner core and the outer skeleton of the bushing. The acquisition unit acquires the vibration displacement data of the bushing in the radial and axial directions at a fixed sampling frequency, which is not less than 5 times the maximum expected vibration frequency of the bushing. The acquired vibration displacement time series data is a discrete time series, denoted as where is the th sampling moment, , is the total number of sampling points, and the sampling time interval is , satisfying . The multi-dimensional corrosion factor sequence data includes corrosion parameters in 5 dimensions: environmental temperature, relative humidity, sodium chloride salt spray concentration, sulfur dioxide concentration, and ultraviolet radiation intensity. It is obtained by multiple environmental acquisition units arranged around the bushing. Each corrosion factor corresponds to an independent time series data, and the sampling frequency can be the same or different from that of the vibration displacement time series data. Denote the time series data of the th corrosion factor as where is the th corrosion factor sampling moment, , is the total number of corrosion factor sampling points, , is the total number of dimensions of the corrosion factors.

[0020] In the time dimension, based on the vibration period in the vibration displacement time series data, resample the multi-dimensional corrosion factor sequence data to obtain a time-aligned corrosion factor sequence. Perform a discrete Fourier transform on the vibration displacement time series data and extract the main frequency component in the frequency domain. The calculation process of the discrete Fourier transform satisfies the formula:

[0021] where is the vibration displacement time series data at frequency The amplitude in the frequency domain at that point, For the first Each frequency point, , The imaginary unit, The frequency point corresponding to the maximum amplitude in the frequency domain is extracted as the dominant frequency component, denoted as . The duration of a single vibration period is calculated based on the dominant frequency component, and the calculation process satisfies the formula:

[0022] in The duration of a single vibration cycle is used as the time window. The multidimensional corrosion factor sequence data is divided into multiple corrosion factor subsequences, with each subsequence having a duration consistent with the time window length, and no overlap between adjacent time windows. The arithmetic mean of each corrosion factor subsequence within its corresponding time window is calculated, and this mean is used as the corrosion factor value at the center of the time window. This calculation is performed sequentially for all time windows to generate a time-aligned corrosion factor sequence. Each sampling moment of the time-aligned corrosion factor sequence corresponds to the center moment of the vibration cycle in the vibration displacement time series data, ensuring a unified time reference for the corrosion factor data and the vibration displacement data.

[0023] In the spatial dimension, corrosion factor data from the time-aligned corrosion factor sequence are mapped to corresponding spatial coordinate nodes in the 3D digital twin mesh model of the suspension damping bushing. The 3D digital twin mesh model is obtained by acquiring the geometric contour of the bushing entity through 3D topographic scanning. After generating the 3D solid model, a finite element mesh generation tool is used to generate the mesh, which is a tetrahedral mesh. Each spatial coordinate node corresponds to a unique coordinate in 3D space. Each node is associated with material property parameters of the bushing, including elastic modulus, Poisson's ratio, material permeability, initial thickness of the anti-corrosion layer, and initial adhesion strength. To clearly illustrate the node property definitions of the mesh model, the following table is provided: Table 1. Spatial coordinate node attribute definition table for the three-dimensional digital twin mesh model of suspension damping bushing.

[0024] Note: All node attribute parameters were obtained through 3D scanning and material performance testing to ensure that the geometric and physical properties of the mesh model and the solid bushing are consistent.

[0025] The physical spatial coordinates of multiple environmental acquisition units arranged around the suspension damping bushing and the raw corrosion factor data collected by these units are obtained. The environmental acquisition units are arranged to cover different orientations of the bushing in the axial, radial, and circumferential directions, and the number of acquisition units is no less than four. Using an inverse distance weighted interpolation algorithm, with the physical spatial coordinates of the environmental acquisition units as interpolation nodes and the raw corrosion factor data as interpolation attributes, the interpolated corrosion factor data at each spatial coordinate node in the 3D digital twin mesh model is calculated. The calculation process satisfies the following formula:

[0026] in In the three-dimensional digital twin mesh model, the first Interpolated corrosion factor data at each spatial coordinate node This represents the total number of environmental data acquisition units. For the first Raw corrosion factor data collected by each environmental acquisition unit. For the first The spatial coordinate node and the first The Euclidean distance between the physical spatial coordinates of each environmental acquisition unit. The power exponent is the inverse distance weighted exponent, with a value ranging from 1 to 3, preferably 2. The interpolated erosion factor data at the same time are assigned to the corresponding spatial coordinate nodes, completing the grid mapping of the erosion factor data in the spatial dimension.

[0027] Generate a spatiotemporally aligned multiphysics fusion dataset. Extract the stress response characteristics of each spatial coordinate node in the 3D digital twin mesh model corresponding to the vibration displacement time series data. Input the peak value of the vibration displacement time series data as the displacement boundary condition into the 3D digital twin mesh model. Calculate the Mises equivalent stress amplitude of each spatial coordinate node in each vibration cycle using a finite element solver. This value is taken as the stress response characteristic of that node and denoted as... ,in This refers to the node number in the spatial coordinate system. Dimensionless processing is performed on the stress response characteristics and corrosion factor data mapped to the same spatial coordinate node to obtain the normalized stress characteristic vector and the normalized corrosion factor vector. The normalization of the stress response characteristics satisfies the following formula:

[0028] in For the first Normalized stress characteristics of each spatial coordinate node, This represents the minimum equivalent stress amplitude across all spatial coordinate nodes. This represents the maximum equivalent stress amplitude across all spatial coordinate nodes. The normalization of corrosion factor data satisfies the formula:

[0029] in For the first At the nth spatial coordinate node The normalized values ​​of each corrosion factor. For the first At the nth spatial coordinate node Interpolated data for each corrosion factor For all spatial coordinate nodes, the first The minimum value of each corrosion factor, For all spatial coordinate nodes, the first The maximum value of the corrosion factor. The normalized stress eigenvector and the normalized corrosion factor vector are concatenated along the eigendimensional axis to obtain the multi-physics node eigenvector of the spatial coordinate node, the th... The feature vector of each node is denoted as . The multiphysics fusion dataset is composed of the feature vectors of all spatial coordinate nodes.

[0030] A multiphysics fusion dataset is input into a pre-constructed digital twin model. The graph neural network within the digital twin model extracts the correlation between stress transfer characteristics and corrosion medium diffusion characteristics among mesh nodes in the 3D digital twin mesh model. A graph topology is constructed, with spatial coordinate nodes in the 3D digital twin mesh model as graph nodes and the physical connections between adjacent spatial coordinate nodes as graph edges. Adjacent spatial coordinate nodes refer to nodes sharing a unit face in the mesh model. The adjacency matrix of the graph topology is... for sparse matrix, where Let be the total number of nodes in the graph. If the number of nodes is 1, then the total number of nodes in the graph is With nodes If a physical connection exists, then the corresponding element in the adjacency matrix... ,otherwise Based on the spatial distance between adjacent graph nodes and the material permeability parameter, the initial physical weights of the graph edges are calculated. The calculation process satisfies the following formula:

[0031] in For graph nodes Graph Nodes The initial physical weights of the graph edges between them. For graph nodes Graph Nodes The equivalent material permeability between the two nodes is taken as the arithmetic mean of the material permeability of the two nodes. For graph nodes Graph Nodes The Euclidean distance between them. Graph neural networks, based on graph topology and initial physical weights, transmit node features from a multiphysics fusion dataset along graph edges through graph convolutional layers, extracting the correlation between stress transmission features and corrosion medium diffusion features. The computation process of the graph convolutional layer satisfies the formula:

[0032] in For the first The output feature matrix of the layer graph convolutional layer, For the first The input feature matrix of the layer-by-layer convolutional layer, the initial input This is the node feature matrix of the multiphysics fusion dataset. To add self-loops to the adjacency matrix, It is the identity matrix. for The degree matrix satisfies , For the first The trainable weight matrix of a layered graph convolutional layer This is a linear rectified activation function. Through feature transfer and aggregation in multiple graph convolutional layers, the output feature matrix contains the coupled correlation features between stress transfer and corrosion medium diffusion between nodes, that is, the correlation between stress transfer features and corrosion medium diffusion features.

[0033] Based on the correlation, the degradation status data of the anti-corrosion layer of the suspension damping bushing at different spatial locations are output. The feature matrix output from the graph convolutional layer corresponding to the correlation is input into the fully connected layer of the digital twin model to calculate the current thickness attenuation and adhesion strength attenuation of the anti-corrosion layer at each spatial coordinate node. The calculation process of the fully connected layer satisfies the formula:

[0034] in For the first The output vector of the fully connected layer at each node contains two elements: the current thickness decay amount and the value of the current thickness decay. and adhesion strength attenuation , This is the weight matrix of the fully connected layer. This is the bias vector of the fully connected layer. For the first The output of the layer graph convolutional layer is the first The feature vectors of each node. The current thickness attenuation and adhesion strength attenuation are input into a preset micro-damage mapping function to calculate the microcrack propagation probability at the spatial coordinate nodes. The calculation process of the micro-damage mapping function satisfies the formula:

[0035] in For the first The probability of microcrack propagation at each node. The damage coefficient corresponding to thickness attenuation. This represents the damage coefficient corresponding to the decrease in adhesion strength. This is the initial thickness of the anti-corrosion layer. The initial adhesion strength of the anti-corrosion layer is defined. Based on the microcrack propagation probability and a preset degradation threshold range, the degradation level of the anti-corrosion layer corresponding to each spatial coordinate node is determined. The preset degradation threshold range is determined based on the test data of the anti-corrosion layer material performance, and the range is divided as follows: Corresponding to Level 1 Degeneration, Corresponding to level 2 degradation, Corresponding to level 3 degradation, Corresponding to level 4 degradation, Corresponding to 5 levels of degradation. The degradation level of the anti-corrosion layer, the corresponding thickness attenuation, the adhesion strength attenuation, and the microcrack propagation probability of all spatial coordinate nodes are output as anti-corrosion layer degradation state data.

[0036] In this embodiment, the temporal reference of vibration data and corrosion data is aligned through resampling in the time dimension, and the spatial coordinates of corrosion factor data and structural mechanics grid nodes are aligned through grid mapping in the spatial dimension. The generated multiphysics fusion dataset provides the graph neural network with coupled feature inputs under the same spatiotemporal reference. The graph neural network extracts the correlation features between stress transmission and corrosion diffusion based on the topological connection relationship of grid nodes, so that the model calculation process fits the physical mechanism of stress and corrosion coupling. The output anti-corrosion layer degradation state data is consistent with the physical evolution process of the bushing during actual service.

[0037] Please refer to Figure 2 In a preferred embodiment, to address the vibration dominant frequency shift caused by dynamic changes in vehicle driving conditions, the time-frequency analysis and resampling process of the vibration displacement time series data is refined. The vibration displacement time series data is segmented into multiple sliding time windows with overlapping regions, and the window length of each sliding time window is [missing information]. There are sampling points, and the length of the overlapping region between adjacent sliding time windows is . The overlap rate should be no less than 50%, preferably 75%, to ensure the temporal continuity of the time-frequency analysis. A short-time Fourier transform is performed on the data for each sliding time window to obtain the time-frequency spectrum. The calculation process of the short-time Fourier transform follows the formula:

[0038] in For a moment ,frequency The short-time Fourier transform amplitude at that point. The window function is preferred, either the Hanning window or the Hamming window. The center moment of the sliding time window. Let be the frequency variable. In the time-spectrum graph, the frequency change trajectory corresponding to the energy peak is traced along the time axis. The frequency corresponding to the energy peak within the current sliding time window is determined as the dynamic dominant frequency at the current moment, denoted as . ,in For the first The center time of each sliding time window. The dynamic main frequency within adjacent sliding time windows is smoothed and filtered to eliminate random noise interference during the main frequency extraction process. The calculation process of the smoothing filter satisfies the formula:

[0039] in For the first The dynamic main frequency after filtering at the center moment of a sliding time window. This is the half-window length for smoothing filtering, with a value ranging from 2 to 5. This represents the dynamic dominant frequency value of adjacent sliding time windows. The duration of a single dynamic vibration period, varying with time, is calculated from the reciprocal of the filtered dynamic dominant frequency. The calculation process satisfies the formula:

[0040] in For the first The duration of a single dynamic vibration cycle corresponding to the center moment of each sliding time window.

[0041] Using the duration of the dynamic single vibration cycle corresponding to each moment as a time window, dynamic resampling processing is performed on the multidimensional corrosion factor sequence data. For each moment... Centered on that moment, with Given a time window length, extract corrosion factor sampling data within that time window, calculate the arithmetic mean of the corrosion factor data within that window, and use this average as the time interval. The corrosion factor values ​​at each location are calculated sequentially at the center time of all sliding time windows to generate a corrosion factor sequence aligned with the dynamic vibration period. For boundary regions with insufficient sampling points within a time window, linear interpolation is used to supplement the corrosion factor sampling data before mean calculation. To clearly illustrate the parameter configurations for the time-frequency analysis and resampling process, the following table is provided: Table 2. Time-frequency analysis and resampling parameter configuration table for vibration displacement time series data

[0042] In this embodiment, the dynamic dominant frequency that changes with time is extracted by the short-time Fourier transform of the sliding time window. Combined with smoothing filtering, random noise interference in the dominant frequency extraction process is eliminated. Based on the dynamic dominant frequency, the dynamic vibration period that changes with the working conditions is calculated. The corrosion factor sequence is resampled based on the dynamic vibration period, which achieves high-precision time alignment between vibration data and corrosion data under varying working conditions. This avoids the alignment error of fixed-period resampling when the working conditions fluctuate, and further ensures the spatiotemporal consistency of the data input to the model.

[0043] refer to Figure 3 In a preferred embodiment, to address the corrosion factor interpolation error caused by the curved geometry of the suspension damping bushing, the process of mapping corrosion factor data in the spatial dimension and generating the multi-physics fusion dataset is refined. The physical spatial coordinates of multiple environmental acquisition units arranged around the bushing, as well as the original corrosion factor data collected by each unit, are acquired. The environmental acquisition units are positioned to cover the bushing's axial ends, radial up and down, and circumferential directions. The number of acquisition units is no less than six to ensure omnidirectional coverage of the three-dimensional spatial interpolation. The initial Euclidean distance between the physical spatial coordinates of the environmental acquisition units and the spatial coordinate nodes to be interpolated in the three-dimensional digital twin mesh model is calculated, satisfying the formula:

[0044] in For the first The spatial coordinate node and the first The initial Euclidean distance between each environmental acquisition unit For the first The three-dimensional coordinates of each spatial coordinate node. For the first The physical space three-dimensional coordinates of each environmental acquisition unit are obtained. The surface curvature data of the three-dimensional digital twin mesh model along the initial Euclidean distance line is acquired. The surface curvature is the Gaussian curvature of the bushing's outer surface at the intersection of this line and the surface, calculated using differential geometry of the three-dimensional mesh model. The Gaussian curvature is the product of the two principal curvatures at that point, denoted as... ,in Corresponding to convex surfaces, Corresponding to concave surfaces, Corresponding plane. The initial Euclidean distance is corrected using surface curvature data to obtain the surface correction distance. The calculation process satisfies the formula:

[0045] in This is the distance after surface correction. This is the curvature correction coefficient, with a value ranging from 0.1 to 1.0, preferably 0.5. The surface curvature is the absolute value of the Gaussian curvature; the greater the surface curvature, the longer the surface correction distance. The surface correction distance is used instead of the initial Euclidean distance as the distance parameter in the inverse distance weighted interpolation algorithm to calculate the interpolated corrosion factor data.

[0046] refer to Figure 4 The generation process of the multiphysics fusion dataset is refined. The stress response characteristics of each spatial coordinate node in the 3D digital twin mesh model are extracted from the vibration displacement time-series data. The peak vibration displacement corresponding to each time-aligned vibration cycle is used as the displacement boundary condition and input into the 3D finite element mesh model. Using the linear elastic finite element analysis method, the Mises equivalent stress amplitude, maximum principal stress amplitude, and stress cycle number of each spatial coordinate node within that vibration cycle are obtained. These parameters constitute the stress response feature vector of that node, denoted as... ,in The amplitude of the Mises equivalent stress. The maximum principal stress amplitude, The stress cycle number is used. Dimensionless processing is performed on the stress response features and corrosion factor data mapped to the same spatial coordinate node. A minimum-maximum normalization method is used to normalize each parameter in the stress response feature vector and each parameter in the corrosion factor vector, resulting in a normalized stress feature vector and a normalized corrosion factor vector. The normalized stress feature vector and the normalized corrosion factor vector are concatenated along their feature dimensions to obtain the multi-physics node feature vector of the spatial coordinate node. If the dimension of the normalized stress feature vector is... The dimension of the normalized corrosion factor vector is Then the dimension of the concatenated multiphysics node feature vector is The multiphysics node eigenvectors of all spatial coordinate nodes are arranged in order of node number, forming a dimension of A multiphysics fusion dataset, in which This represents the total number of spatial coordinate nodes in the 3D digital twin mesh model. The following table is provided to clearly illustrate the parameter configuration for corrosion factor interpolation calculation: Table 3. Interpolation Parameters and Curvature Correction Coefficients for Multidimensional Corrosion Factor

[0047] Note: The curvature correction factor is adjusted according to the geometric features of the bushing surface. Higher correction factors are used for areas with large curvature, such as rounded corners and steps.

[0048] In this embodiment, by introducing bushing surface curvature data to correct the distance parameter of the inverse distance weighted interpolation, the spatial interpolation result of the corrosion factor is made to fit the geometric shape of the bushing surface and the actual diffusion path of the corrosive medium, thus improving the accuracy of spatial mapping. By refining the multi-dimensional extraction of stress response features and the normalization and splicing of feature vectors, the generated multi-physics fusion dataset contains structural mechanical features and environmental corrosion features under the same spatiotemporal benchmark, providing complete coupled feature inputs for subsequent feature extraction of graph neural networks, further ensuring the accuracy of model calculation results.

[0049] refer to Figure 5 In a preferred embodiment, to address the issues of local feature loss during graph neural network feature extraction and the inability to reflect the time accumulation effect during the corrosion layer degradation process, the graph neural network structure in the digital twin model and the calculation process of the corrosion layer degradation state are refined. A graph topology structure is constructed, with spatial coordinate nodes in the 3D digital twin mesh model as graph nodes and the physical connection relationships between adjacent spatial coordinate nodes as graph edges. The adjacency matrix of the graph topology structure... for sparse matrix, where Let be the total number of nodes in the graph. If the number of nodes is 1, then the total number of nodes in the graph is With nodes In a mesh model, if there are physical connections, i.e., faces or edges sharing the same mesh cell, then the corresponding elements in the adjacency matrix... ,otherwise Based on the spatial distance between adjacent graph nodes and the material permeability parameter, the initial physical weights of the graph edges are calculated. For nodes with connectivity relationships... With nodes The equivalent material permeability between the two nodes is calculated. The equivalent material permeability is taken as the harmonic average of the permeabilities of the materials to which the two nodes belong, in order to conform to the diffusion resistance characteristics of the corrosive medium at different material interfaces. The calculation process satisfies the formula:

[0050] in For nodes The permeability of the material For nodes The permeability of the material is considered. Combined with the spatial distance between nodes, the initial physical weights of the graph edges are calculated. These initial physical weights reflect the ease of diffusion of the corrosive medium between adjacent nodes and the efficiency of stress transfer between adjacent nodes, providing a weighted basis consistent with physical mechanisms for feature transfer in graph convolution.

[0051] A dilated graph convolutional layer with multiple dilation rates is used to process node features in the multiphysics fusion dataset in parallel. Different dilation rates correspond to different hop counts for aggregating neighboring graph node features. For dilation rates of... The dilated graph convolution, whose feature aggregation neighborhood is the node Skip neighborhood, that is, starting from the target node and traversing... The expansion rate of the neighborhood formed by all nodes reachable by the edges of the graph. The corresponding 1-hop neighborhood, i.e., directly adjacent nodes; expansion rate. The corresponding 2-hop neighborhood, i.e., the neighboring nodes of adjacent nodes; expansion rate This corresponds to a 3-hop neighborhood. For each dilation rate of the expanded graph convolutional layer, graph convolution is performed to obtain the local aggregated features within the corresponding neighborhood range. The calculation process satisfies the formula:

[0052] in expansion rate The corresponding local aggregated features output by the dilated graph convolutional layer For adding self-loop Skip adjacency matrix for Skip adjacency matrix, where If and only if node With nodes The shortest path length between them is , for The degree matrix, expansion rate The corresponding convolutional layer can be trained with weight matrices. The local aggregated features output from convolutional layers with different dilation rates are concatenated along the feature dimension to obtain a multi-scale aggregated feature matrix. The calculation process satisfies the formula:

[0053] in This is the concatenated multi-scale aggregated feature matrix. For multiple preset expansion rates, the preferred one is... , This involves a feature concatenation operation. The concatenated local aggregated features are input into the residual network layer. The residual network layer adds the concatenated local aggregated features to the initial node features in the multiphysics fusion dataset, resulting in the global association features output by the graph convolutional layer. The calculation process satisfies the following formula:

[0054] in The global correlation features output by the residual network layer. This is the initial node feature matrix of the multiphysics fusion dataset. The weight matrix for the residual mapping is used to map the initial node features to the same feature dimension as the multi-scale aggregated features.

[0055] refer to Figure 6 The global correlation features output by the graph neural network are input into the fully connected layer of the digital twin model to calculate the current thickness attenuation and adhesion strength attenuation of the anti-corrosion layer at each spatial coordinate node. The fully connected layer contains two linear transformation layers and one nonlinear activation layer, and the two output parameters are the thickness attenuation. With adhesion strength attenuation ,in This represents the current computation time. The current thickness attenuation, adhesion strength attenuation, and the microcrack propagation probability sequence of the spatial coordinate nodes at historical moments are input into the Long Short-Term Memory (LSTM) network. The LTM network's forgetting gate discards historical microcrack propagation probabilities irrelevant to the current stress state. The LTM network's input gate combines the current thickness attenuation and adhesion strength attenuation to update the cell state. The updated cell state is then output through the output gate as the output probability of the dynamically adjusted micro-damage mapping function. The input to the LTM network includes the input vector at the current moment. The hidden state of historical moments Cellular state at a historical moment The hidden state of historical moments It contains information about the probability sequence of microcrack propagation at historical moments.

[0056] The calculation process of the forgetting gating unit satisfies the formula:

[0057] in The output of the forget gate has a value range of 0 to 1. Here is the weight matrix for the forget gate. Let be the bias vector of the forget gate. This involves concatenating the historical hidden state with the current input vector. The activation function is sigmoid; the closer the output value is to 0, the higher the degree of discarding of corresponding historical information. The calculation process of the input gating unit satisfies the formula:

[0058]

[0059] in The output of the input gate has a value range of 0 to 1. Here is the weight matrix of the input gate. Let be the bias vector of the input gate. Candidate cell state, This is the weight matrix for the candidate cell states. This is the bias vector for the candidate cell state. The hyperbolic tangent activation function is used. The cell state update process satisfies the following formula:

[0060] in This is the updated cell state at the current moment. To preserve historical cell state information, This represents the newly added cell state information at the current moment. The calculation process for the output gating unit and the microcrack propagation probability satisfies the following formula:

[0061]

[0062]

[0063] in The output of the output gate has a value range of 0 to 1. This is the weight matrix of the output gate. This is the bias vector for the output gate. The current hidden state. This is the weight matrix of the output layer. This is the bias vector for the output layer. For the first Each node at the current moment The probability of microcrack propagation.

[0064] Based on the microcrack propagation probability and a preset degradation threshold range, the degradation level of the anti-corrosion layer to which each spatial coordinate node belongs is determined. The preset degradation threshold range is determined based on the material performance test data of the anti-corrosion layer and industry standards. Different degradation levels correspond to different maintenance strategies. The degradation level of the anti-corrosion layer, the corresponding thickness attenuation, the adhesion strength attenuation, and the microcrack propagation probability of all spatial coordinate nodes are output as anti-corrosion layer degradation state data. Simultaneously, the degradation level can be mapped to a 3D digital twin mesh model to generate a visual cloud map of the anti-corrosion layer degradation state. To clearly illustrate the network structure parameter configuration of the digital twin model, the following table is provided: Table 4. Parameter Configuration Table for Layer Structure of Graph Neural Network and Long Short-Term Memory Network in Digital Twin Model

[0065] In this embodiment, multi-scale node feature aggregation is achieved through multi-expansion rate dilation graph convolutional layers. Combined with residual network layers, the key physical features of the initial nodes are preserved. The extracted correlation between stress transmission and corrosion diffusion features includes coupled local and global physical information. A dynamic micro-damage mapping function is constructed through a long short-term memory network, enabling the calculation of microcrack propagation probability to reflect the time-cumulative effect of anti-corrosion coating degradation. The output anti-corrosion coating degradation status data can accurately reflect the dynamic degradation process of the bushing during service, further improving the prediction accuracy and reliability of the digital twin model.

Claims

1. A digital twin analysis method for the corrosion resistance of suspension damping bushings, characterized in that, include: Obtain the vibration displacement time series data of the suspension damping bushing during service and the multidimensional corrosion factor sequence data of the environment in which it is located; In the time dimension, the multidimensional corrosion factor sequence data is resampled based on the vibration period in the vibration displacement time series data to obtain a time-aligned corrosion factor sequence. In the spatial dimension, the corrosion factor data in the time-aligned corrosion factor sequence is mapped to the corresponding spatial coordinate nodes in the three-dimensional digital twin mesh model of the suspension damping bushing to generate a spatiotemporally aligned multiphysics fusion dataset. The multiphysics fusion dataset is input into a pre-constructed digital twin model. The graph neural network in the digital twin model is used to extract the correlation between stress transfer characteristics and corrosion medium diffusion characteristics between grid nodes in the three-dimensional digital twin mesh model. Based on the correlation, the degradation status data of the anti-corrosion layer of the suspension damping bushing at different spatial locations are output.

2. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 1, characterized in that, The step of resampling the multidimensional corrosion factor sequence data based on the vibration period in the vibration displacement time series data includes: performing a Fourier transform on the vibration displacement time series data to extract the main frequency component in the frequency domain signal. The duration of a single vibration cycle is calculated based on the dominant frequency component. Using the duration of a single vibration cycle as a time window, the multidimensional corrosion factor sequence data is divided into multiple corrosion factor subsequences; Calculate the mean of each corrosion factor subsequence within the corresponding time window, and use the mean as the corrosion factor value at the center of the time window to generate the time-aligned corrosion factor sequence.

3. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 1, characterized in that, The step of mapping the corrosion factor data in the time-aligned corrosion factor sequence to the corresponding spatial coordinate nodes in the three-dimensional digital twin mesh model of the suspension damping bushing includes: acquiring the physical spatial coordinates of multiple environmental sensors arranged around the suspension damping bushing and the raw corrosion factor data collected by the environmental sensors. Using the inverse distance weighted interpolation algorithm, with the physical spatial coordinates of the environmental sensor as the interpolation nodes and the original corrosion factor data as the interpolation attribute, the interpolated corrosion factor data at each spatial coordinate node in the three-dimensional digital twin mesh model is calculated. The interpolated corrosion factor data at the same time are assigned to the corresponding spatial coordinate nodes.

4. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 1, characterized in that, The generation of the spatiotemporally aligned multiphysics fusion dataset includes: extracting the stress response features corresponding to each spatial coordinate node in the three-dimensional digital twin mesh model in the vibration displacement time series data; The stress response characteristics and the corrosion factor data mapped to the same spatial coordinate node are subjected to dimensionless processing to obtain a normalized stress feature vector and a normalized corrosion factor vector. The normalized stress feature vector and the normalized corrosion factor vector are concatenated along the feature dimension to obtain the multi-physics node feature vector of the spatial coordinate node. The multiphysics fusion dataset is composed of the feature vectors of all the spatial coordinate nodes.

5. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 1, characterized in that, The step of extracting the correlation between stress transfer characteristics and corrosion medium diffusion characteristics between grid nodes in the three-dimensional digital twin mesh model through the graph neural network in the digital twin model includes: constructing a graph topology structure with spatial coordinate nodes in the three-dimensional digital twin mesh model as graph nodes and physical connection relationships between adjacent spatial coordinate nodes as graph edges; The initial physical weights of the graph edges are calculated based on the spatial distance between adjacent graph nodes and the material permeability parameter. The graph neural network, based on the graph topology and the initial physical weights, extracts the association relationship by passing node features from the multiphysics fusion dataset to the graph edges through graph convolutional layers.

6. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 1, characterized in that, The step of outputting the corrosion degradation status data of the suspension damping bushing at different spatial locations according to the correlation includes: inputting the correlation into the fully connected layer of the digital twin model, and calculating the current thickness attenuation and adhesion strength attenuation of the corrosion layer at each spatial coordinate node; The current thickness attenuation and the adhesion strength attenuation are input into a preset micro-damage mapping function to calculate the microcrack propagation probability at the spatial coordinate node. Based on the microcrack propagation probability and the preset degradation threshold range, the degradation level of the anti-corrosion layer to which each spatial coordinate node belongs is determined, and the degradation level of the anti-corrosion layer of all spatial coordinate nodes is output as the degradation status data of the anti-corrosion layer.

7. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 2, characterized in that, The extraction of the dominant frequency component from the frequency domain signal includes: dividing the vibration displacement time series data into multiple sliding time window data with overlapping regions; Perform a short-time Fourier transform on each of the sliding time window data to obtain a time-spectrum graph; In the time-frequency spectrum, the frequency change trajectory corresponding to the energy peak is traced along the time axis, and the frequency corresponding to the energy peak within the current sliding time window is determined as the dynamic dominant frequency at the current moment. The step of calculating the duration of a single vibration cycle based on the dominant frequency component includes: performing a smoothing filter on the dynamic dominant frequency within adjacent sliding time windows, and calculating the duration of a dynamic single vibration cycle that changes with time based on the reciprocal of the filtered dynamic dominant frequency.

8. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 3, characterized in that, The step of using the inverse distance weighted interpolation algorithm to calculate the interpolated corrosion factor data at each spatial coordinate node in the three-dimensional digital twin mesh model includes: calculating the initial Euclidean distance between the physical spatial coordinates of the environmental sensor and the spatial coordinate node; Obtain the surface curvature data of the three-dimensional digital twin mesh model along the direction of the initial Euclidean distance line; The initial Euclidean distance is corrected using the surface curvature data to obtain the surface correction distance, wherein the greater the surface curvature, the longer the surface correction distance. The surface correction distance is used to replace the initial Euclidean distance as the distance parameter in the inverse distance weighted interpolation algorithm to calculate the interpolated corrosion factor data.

9. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 5, characterized in that, The step of passing node features in the multiphysics fusion dataset through graph convolutional layers on graph edges includes: using an expanded graph convolutional layer with multiple expansion rates to process node features in the multiphysics fusion dataset in parallel, wherein different expansion rates correspond to different hop counts of adjacent graph node feature aggregation ranges. The local aggregated features output from the convolutional layers of the expanded maps with different expansion rates are stitched together; The spliced ​​local aggregated features are input into the residual network layer, and the residual network layer adds the spliced ​​local aggregated features to the initial node features in the multiphysics fusion dataset to obtain the global association features output by the graph convolutional layer.

10. The digital twin analysis method for the corrosion resistance of suspension damping bushings according to claim 6, characterized in that, The step of inputting the current thickness attenuation and the adhesion strength attenuation into a preset micro-damage mapping function includes: inputting the current thickness attenuation, the adhesion strength attenuation, and the microcrack propagation probability sequence of the spatial coordinate node at a historical moment into a long short-term memory network; The forgetting gating unit of the long short-term memory network discards the historical microcrack propagation probability that is irrelevant to the stress state at the current moment. The cell state is updated by combining the current thickness decay amount and the adhesion strength decay amount at the current moment using the input gating unit of the long short-term memory network; The updated cell state is output through the output gating unit as the output probability of the dynamically adjusted micro-damage mapping function.

Images