Oil bearing load capacity evaluation method based on virtual simulation

Abstract

The application relates to an oil bearing load capacity evaluation method based on virtual simulation, which comprises the following steps: obtaining working condition data samples of an oil bearing under complex working conditions of composite loads in multiple typical operating states, and identifying bearing structure partitions to establish a working condition partition mapping table; establishing a multi-physical field sensitive area identification model for each partition, and dynamically adjusting a global finite element model and a local high-precision micro-element sub-model by using a multi-physical field coupling calibration factor; and extracting time sequence characteristics from the multi-physical field joint simulation result after dynamic parameter adjustment, and obtaining a multi-dimensional parameter sequence related to a typical failure mechanism of the oil bearing.

Description

Technical Field

[0001] This application relates to virtual simulation of workpiece operating conditions, specifically to a method for evaluating the load-bearing capacity of oil-impregnated bearings based on virtual simulation. Background Technology

[0002] Currently, the assessment of the load-bearing capacity and durability analysis of oil-impregnated bearings is gradually developing towards high-fidelity virtual simulation and multiphysics coupling, especially in the fields of aerospace, high-end manufacturing, and precision machinery. This places higher technical demands on the engineering reliability, failure mechanism identification, and life prediction of oil-impregnated bearings under complex, dynamic, and combined load conditions. Mainstream technologies are mostly based on global finite element analysis, supplemented by certain contact mechanics and lubrication theories, using simulation methods to calculate parameters of single physical fields such as structural stress, temperature, and lubrication state. However, facing the complex and variable working conditions of bearings in actual operation, such as simultaneous dynamic loads, impact loads, asymmetric and superimposed loads, as well as the highly coupled dynamic processes between multiple physical fields such as lubrication, structure, and heat, existing methods mainly have the following technical limitations and shortcomings:

[0003] Currently, international virtual simulations for complex operating conditions of oil-impregnated bearings largely focus on numerical analysis of single-physics fields or weakly coupled multiphysics fields. Common methods include static finite element stress field simulation, classical Reynolds lubrication theory and temperature rise calculation, and weakly coupled finite element-fluid dynamics methods. These methods generally employ preset load histories or typical single operating conditions, such as constant loads or simplified dynamic loads, to qualitatively or quantitatively analyze the stress distribution, oil film thickness, and temperature field state of the bearing. Some advanced techniques attempt to introduce fluid-solid (fluid-structure interaction) or solid-thermal (thermo-structure interaction) coupling into the analysis, but due to limitations such as computational cost, data synchronization interfaces, or scale differences, they generally suffer from bottlenecks in overall resolution and real-time multiphysics interaction, making it difficult to adapt to the constantly changing complex loads and real failure processes in engineering. Furthermore, simulation models often employ fixed parameter assumptions, lacking the ability to accurately capture rapid changes in the physical field and local failure mechanisms under extreme conditions; accuracy mainly relies on engineering experience parameters or subsequent manual corrections.

[0004] In other words, the problems with existing technologies include:

[0005] 1. Due to the limited depth and accuracy of multi-physics coupling, existing technologies struggle to achieve high-speed, accurate, and bidirectional interaction between multi-physics fields (structure, lubrication, and heat) in sub-regions or even micrometer-level sensitive parts within the same simulation process. They often approximate the interaction with single-field superposition or weak coupling, thus overlooking the triggering effect of key coupling mechanisms on extreme failures.

[0006] 2. The conflict between local extreme response and global model resolution: Existing finite element models are limited by overall scale and computing power, and cannot perform targeted super-refined simulations of high-risk local areas. Local dynamics and micro-lubrication and thermal behavior are easily weakened or even ignored.

[0007] 3. Lack of data-driven response calibration and model adaptation mechanisms. Traditional simulations are mostly based on theoretical formulas or static settings, making it difficult to dynamically update parameters and perform intelligent calibration based on actual historical working conditions, experimental observations, and new failure categories. As a result, the applicability and reliability of the model are insufficient, and it is difficult to automatically evolve and expand.

[0008] 4. The ability to reconstruct the failure mechanism under dynamic conditions of composite loads is insufficient. Existing methods fail to systematically reproduce the real physical field evolution process under the superposition of dynamic loads, impacts, asymmetric conditions, etc., making it difficult for failure prediction and diagnosis to cover the entire life cycle and unknown extreme events. Summary of the Invention

[0009] To address the problems existing in the prior art, the purpose of this application is to provide a method for evaluating the load-bearing capacity of oil-impregnated bearings based on virtual simulation.

[0010] In order to solve the above-mentioned technical problems, the present invention provides a method for evaluating the load-bearing capacity of oil-impregnated bearings based on virtual simulation.

[0011] The technical solution of this invention is implemented as follows:

[0012] A method for evaluating the load-carrying capacity of oil-impregnated bearings based on virtual simulation, comprising:

[0013] S1: Obtain working condition data samples of oil-impregnated bearings under multiple typical operating states under complex working conditions of combined loads. The working condition data includes working condition labels such as dynamic load, impact load and asymmetric load, and the bearing structure partitions are identified to establish a working condition partition mapping table.

[0014] S2: For different structural partitions in the working condition partition mapping table, collect multi-physics time series experimental data including lubrication, temperature and mechanical stress, and perform noise reduction and standardization processing on the collected multi-physics time series experimental data to obtain high-quality multi-physics basic data.

[0015] S3: Based on multiphysics field basic data and partition identifiers, under the framework of global finite element model, establish a multiphysics field sensitive area identification model for each partition, and output the multiphysics field sensitive area index and physical variable influence weight of each partition;

[0016] S4: For the identified multiphysics sensitive areas, establish local high-precision multiphysics coupled micro-element models, including lubrication fluid dynamics-structure thermal coupling behavior modeling, and associate the parameters of the multiphysics coupled micro-element models with the sensitive area index of the previous step;

[0017] S5: Input the working condition data sample and the local multiphysics coupled micro-element sub-model into the global finite element model, realize the high-speed communication between global and local multiphysics information through the multi-scale mapping method, and output multi-scale multiphysics interactive data.

[0018] S6: Based on multi-scale multi-physics interaction data, a machine learning model is used to predict and invert the local and global physical responses driven by time-series load parameters, calibrate the results of traditional physical modeling, and generate multi-physics coupling calibration factors.

[0019] S7: Utilize multiphysics field coupling calibration factors to dynamically adjust the parameters of the global finite element model and the local high-precision micro-element sub-model, thereby achieving highly adaptable modeling for extreme and atypical working conditions.

[0020] S8: Extract time-series features from the multiphysics joint simulation results after dynamic parameter adjustment to obtain multidimensional parameter sequences related to typical failure mechanisms such as oil film thickness distribution, fluid pressure, temperature rise and contact stress;

[0021] S9: Determine whether the extracted multidimensional parameter sequence meets the preset durability and failure criteria. If not, automatically optimize the modeling granularity and coupling ratio of the multiphysics sensitive area based on the feedback information to form an adaptive iteration of the model.

[0022] S10: Outputs high-fidelity multiphysics-dynamic load co-simulation analysis results after adaptive optimization, used for predicting the durability and failure mechanism of oil-impregnated bearings under complex dynamic conditions and for authoritative load-bearing capacity assessment.

[0023] The beneficial effects of this invention are as follows:

[0024] 1. Significantly Enhanced High-Fidelity Coupling and Constitutive Reduction Capabilities in Multiphysics. This invention proposes for the first time a collaborative coupling mechanism between structural partition sensitivity identification, multiphysics principal component quantification, and local high-resolution micro-element sub-models and global finite element models. Through spatial clustering and sensitivity factor weight quantification methods, it achieves quantitative screening and localization of the dominant failure factors in various functional areas of bearings, such as lubrication, temperature, and mechanical stress. Compared with existing simulation methods based solely on single-field or weak-field interactions, this invention can dynamically capture abnormal coupling and extreme responses between physical fields, reducing the corresponding error in actual test scenarios from generally over 10% to 2%–5%, greatly enhancing the engineering credibility of the load-bearing capacity assessment results.

[0025] 2. Multi-scale, multi-resolution data interoperability mechanism optimizes simulation performance and efficiency. High-density dynamic load, oil temperature, and stress signals obtained through multi-channel acquisition and event-driven annotation are processed through multi-layer indexing, clustering, and feature extraction of structural partitions / sensitive areas, achieving hierarchical refinement of the simulation model from coarse to fine and from global to local. This mechanism not only improves the physical tracking and resolution capability of critical damage areas in bearings but also significantly reduces the computational resource consumption of full-time, full-domain detailed simulation. The simulation efficiency is improved by more than 50% compared to traditional uniform high-precision discretization methods, balancing efficiency and accuracy requirements.

[0026] 3. Machine learning inversion and adaptive parameter optimization significantly enhance model adaptability and generalization. This invention utilizes machine learning models (such as neural networks and transfer learning) to perform data-driven inversion of local-global multiphysics temporal responses and automatically calibrates traditional physical model parameters, constructing a closed-loop model calibration system that integrates the physical and data domains. By adaptively adjusting the modeling granularity and coupling parameters through feedback errors, it significantly suppresses model transfer distortion and criterion failure induced by unknown extreme conditions or atypical complex loads. Experimental results show that under novel loading conditions, the peak error of the fluid pressure extreme values ​​output by the model simulation is reduced by 40% compared to traditional methods, significantly reducing the risk of misjudgment and missed judgment.

[0027] 4. Multidimensional traceability of failure mechanisms and rich practical assessment criteria. Through the extraction of temporal and spatial multidimensional feature criteria and the classification of typical failure mechanisms, the invention can effectively distinguish various typical damage-causing conditions such as abrupt changes in bearing oil film thickness, increased local temperature rise, and abnormal stress distribution. It outputs a multi-physics parameter database covering both spatial and temporal dimensions, constructing a durability discrimination system that better meets real engineering needs. Compared to previous simulations that only provided simplified ultimate load margins, this invention greatly enhances the diagnostic depth and source tracing capabilities for failure causes and evolution paths, providing a solid data foundation for failure prediction and maintenance recommendations. Attached Figure Description

[0028] Figure 1 This application describes a method for evaluating the load-carrying capacity of oil-impregnated bearings based on virtual simulation. Figure 1 . Detailed Implementation

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

[0030] like Figure 1As shown, this embodiment provides a method for evaluating the load-carrying capacity of oil-impregnated bearings based on virtual simulation, specifically including:

[0031] S1: Obtain working condition data samples of oil-impregnated bearings under multiple typical operating states under complex working conditions of combined loads. The working condition data includes working condition labels such as dynamic load, impact load and asymmetric load, and the bearing structure partitions are identified to establish a working condition partition mapping table.

[0032] S2: For different structural partitions in the working condition partition mapping table, collect multi-physics time series experimental data including lubrication, temperature and mechanical stress, and perform noise reduction and standardization processing on the collected multi-physics time series experimental data to obtain high-quality multi-physics basic data.

[0033] S3: Based on multiphysics fundamental data and partition identifiers, a multiphysics sensitive area identification model is established for each partition within the global finite element model framework, and the multiphysics sensitive area index and physical variable influence weight of each partition are output.

[0034] S4: For the identified multiphysics sensitive areas, establish local high-precision multiphysics coupled micro-element models, including modeling of lubrication fluid dynamics-structure thermal coupling behavior, and associate the parameters of the multiphysics coupled micro-element models with the sensitive area index of the previous step.

[0035] S5: Input the working condition data sample and the local multiphysics coupled micro-element sub-model into the global finite element model, and realize the high-speed communication between global and local multiphysics information through the multi-scale mapping method, and output multi-scale multiphysics interactive data.

[0036] S6: Based on multi-scale multi-physics interaction data, a machine learning model is used to predict and invert the local and global physical responses driven by time-series load parameters, calibrate the results of traditional physical modeling, and generate multi-physics coupling calibration factors.

[0037] S7: Utilize multiphysics coupling calibration factors to dynamically adjust the parameters of the global finite element model and the local high-precision micro-element sub-model, achieving highly adaptable modeling for abrupt extremes and atypical working conditions.

[0038] S8: Extract time-series features from the multiphysics joint simulation results after dynamic parameter adjustment to obtain multidimensional parameter sequences related to typical failure mechanisms such as oil film thickness distribution, fluid pressure, temperature rise, and contact stress.

[0039] S9: Determine whether the extracted multidimensional parameter sequence meets the preset durability and failure criteria. If not, automatically optimize the modeling granularity and coupling ratio of the multiphysics sensitive area based on the feedback information to form an adaptive iteration of the model.

[0040] S10: Outputs high-fidelity multiphysics-dynamic load co-simulation analysis results after adaptive optimization, used for predicting the durability and failure mechanism of oil-impregnated bearings under complex dynamic conditions and for authoritative load-bearing capacity assessment.

[0041] S1: Acquire operating condition data samples of oil-impregnated bearings under complex combined load conditions in multiple typical operating states. The operating condition data includes operating condition labels such as dynamic load, impact load, and asymmetric load, and the bearing structure is labeled to establish an operating condition partition mapping table. Specifically, this includes:

[0042] S1.1 Dynamic load tests were conducted on the target oil-impregnated bearing under typical operating conditions to obtain raw operating condition data covering three types of operating conditions: dynamic load, impact load, and asymmetric load. This data formed an operating condition-load type mapping dataset to ensure the diversity and representativeness of the sample's operating conditions.

[0043] The input data includes the structural information of the target oil-impregnated bearing, the definition of typical operating states, and the types of loads (dynamic loads, impact loads, and asymmetric loads) to be acquired under each operating state.

[0044] A multi-channel high-frequency dynamic load test bench (parameters: loading frequency 0~500Hz, maximum load capacity 10kN), an accelerometer (range ±100g), a strain gauge (range 500με), and a force sensor (accuracy 0.1N) are used to collect data in a coordinated manner to define the driving conditions of the bearing under preset typical operating conditions (such as high-speed rotation, emergency stop and start, and off-center load operation).

[0045] Furthermore, through the programmed loading scheme of the experimental platform (parameters: working condition sequence, amplitude change, impact pulse timing), the dynamic load, impact load and asymmetric load are gradually superimposed and switched, and the original working condition excitation signal is acquired in real time at a set synchronous sampling rate (e.g., 10kHz).

[0046] Using a synchronous acquisition module, time alignment and multi-channel synchronization are performed on multi-source load response signals (acceleration, strain, force). For segments with signal distortion or information loss, sampling point interpolation and effective interval compensation are performed.

[0047] Furthermore, by employing signal preprocessing algorithms (such as bidirectional Butterworth filtering with a cutoff frequency of 200Hz), dynamic baseline correction, and drift correction, background interference and zero-point drift noise are eliminated to obtain a purified original operating condition data stream.

[0048] An automatic operating condition discrimination algorithm (parameters: switching interval criteria, peak threshold) is used to perform real-time segmentation and event-driven labeling of the raw data collected in the experiment, and classify and archive it into the operation status-load type mapping dataset.

[0049] Through the above experiments and signal processing procedures, the original working condition data of the target oil-impregnated bearing under various typical operating conditions, including dynamic load, impact load and asymmetric load, were systematically collected and a structured operating condition-load type mapping dataset was formed. This achieved full-space sample coverage of different working condition combinations, improving the representativeness and diversity of subsequent simulation samples.

[0050] For example, in a test scenario of a certain type of high-speed oil-impregnated turbine bearing, three typical operating states were selected: operating state A (stable speed of 18000 rpm), state B (high-speed start-stop), and state C (radial off-center load of 1000 N at the output end). A three-channel load was applied, with each channel applying a sinusoidal dynamic load with an amplitude of 500 N, a 4000 N impact load with a 20 ms width, and a periodically asymmetrically distributed variable load (maximum amplitude 800 N). Signal acquisition used a TDK450 triaxial accelerometer and an HBMFTP-10 force sensor, with a sampling frequency of 10 kHz and a data synchronization error controlled within 0.05 ms. The acquired operating data was filtered using a Butterworth order 4 filter and zero-point drift correction across the entire range to eliminate baseline noise. A custom operating condition discrimination criterion was used: a signal peak value greater than 3500 N was automatically labeled as an impact condition; a periodic sinusoidal wave with an off-center load amplitude variation greater than 20% was labeled as an asymmetrical load; and the remaining loads were classified as dynamic loads. The final output is a three-dimensional mapping dataset of operating states (A, B, C) and load types (dynamic load, impact load, asymmetric load), with a total of 36,000 frames, fully covering typical and extreme working conditions, laying the foundation for subsequent simulation modeling and data mining.

[0051] S1.2 Based on the original working condition data samples, the event-driven labeling algorithm is used to assign load working condition labels to each time series sampling point, generating a working condition data time series label set, which lays a structured foundation for subsequent data partitioning and feature extraction.

[0052] The input conditions include the multi-channel, high-sampling-rate purified raw operating condition data stream obtained in step S1.1. The data structure includes three signals: acceleration (A(t)), strain (ε(t)), and force (F(t)). The time synchronization accuracy is better than 0.05ms, and the sampling frequency is 10kHz. The raw operating condition data has been initially mapped and classified according to typical operating states (such as A, B, C) and load types (dynamic load, impact load, asymmetric load).

[0053] An event-driven annotation algorithm (parameters: peak threshold, periodicity criterion, amplitude change rate, and switching interval determination condition) is employed to automatically identify dynamic load conditions at the sampling point level. Furthermore, a time window sliding segmentation technique (parameters: window length 100ms, step size 10ms) is used to achieve real-time extraction of event features and multi-category load condition clustering within the original signal segment.

[0054] Furthermore, a multi-dimensional signal discrimination index calculation algorithm is employed to calculate the maximum value P for the signal data within the current time window. max =max(F(t)), mean μ = mean(F(t)), root mean square value

[0055] Time-domain statistics such as standard deviation σ = std(F(t)) and periodic distribution parameters are used to extract event discrimination features for impact, dynamic load and asymmetric load.

[0056] Furthermore, by setting a peak threshold P th，impact Period amplitude difference threshold Δ asym and cyclical indicator C cyc The judgment is as follows: when P max >P th，impact The current sampling segment is labeled as an impact load event; when the signal is highly periodic and Δ asym > Set values ​​are marked as asymmetric loads; the rest that meet the set amplitude and period are marked as dynamic loads.

[0057] A segmented event-driven labeling result merging and error correction algorithm is adopted. Boundary smoothing criteria and logical backtracking are introduced for critical intervals or signal abrupt change segments to eliminate mislabeling and ensure the consistency and coherence of time series labels.

[0058] By using the event-driven annotation algorithm described above, the purified original working condition signal is transformed into a set of working condition data time-series labels labeled according to the accuracy of the sampling points. This enables high-precision automatic annotation of working conditions such as dynamic loads, impact loads, and asymmetric loads, and outputs three-dimensional time-series label structure data of working condition-operating state-sampling point, providing a structured foundation for subsequent structural partitioning mapping and multiphysics simulation feature extraction.

[0059] For example, in the raw dataset (sampling frequency 10kHz, total number of frames 36000) obtained for a TDK450 triaxial accelerometer and an HBM FTP-10 force sensor, a peak threshold P is used. th，impact =3500N, period amplitude criterion Δ asym =20, 100ms sliding window and 10ms step window are used for event-driven analysis. For the signal window segment, if max(F(t))>3500N, it is directly marked as an impact condition;

[0060] If the periodicity is significant and the local amplitude deviation exceeds 20%, it is labeled as an asymmetric load condition; otherwise, it is treated as a dynamic load. A time-series label merging and 20ms boundary smoothing rule are used to identify and correct abrupt changes and boundary segments. The final output is a set of time-series labels for each operating state of the oil-impregnated bearing, labeled according to signal points, achieving high consistency and accurate labeling of 36,000 sampled data points without omissions.

[0061] Test results show that the annotation consistency rate reached 99.5%, laying a solid foundation for subsequent structural mapping and simulation feature attribution.

[0062] S1.3 executes a partitioning algorithm on the three-dimensional CAD structural model of the oil-impregnated bearing, subdividing the bearing as a whole into functional areas (such as rolling element area, cage area, lubrication channel area, etc.) and obtaining the structural partition index of each area.

[0063] S1.4 performs high-dimensional data fusion between the time-series label set of working condition data and the structural partition index, and uses a multi-label mapping strategy to generate an integrated working condition-structure mapping table for each structural partition under various load conditions, thus building a correlation foundation for subsequent multi-physics partition identification.

[0064] S1.5 uses an integrated working condition-structure mapping table to perform data consistency verification and redundancy filtering on typical structural partition samples under various operating conditions. Through data cleaning strategies, a high-quality working condition partition mapping table is obtained, providing support for multi-physics field basic data acquisition and model regionalization analysis.

[0065] S2: For different structural partitions in the working condition partition mapping table, collect multi-physics time-series experimental data including lubrication, temperature, and mechanical stress, and perform noise reduction and standardization processing on the collected multi-physics time-series experimental data to obtain high-quality multi-physics basic data. Specifically, this includes:

[0066] S2.1 performs a partition identification operation on the structural partitions in the working condition partition mapping table, extracts the acquisition units of each structural partition based on the partition identifier, and uses them as the target objects for physical field signal acquisition; through partition identification, it ensures that the subsequent physical field data sampling is targeted and highly representative in the working condition space, and outputs a list of structural partition acquisition units.

[0067] S2.2 For the list of acquisition units in the structural partition, multi-channel sensor fusion technology is applied to acquire the time series raw experimental data of lubrication, temperature and mechanical stress in the partition respectively, so as to obtain a dynamic data stream with high coupling of multi-physics fields and form a multi-channel multi-physics raw dataset.

[0068] S2.3 uses a time-frequency domain adaptive filtering algorithm to systematically denoise the original multi-channel multi-physics dataset to eliminate environmental interference and sensor noise, improve the signal-to-noise ratio of each physical field signal, and output the denoised multi-physics clean dataset.

[0069] S2.4 Based on the physical field clean dataset, normalization and standardization transformations are performed on the physical quantities of each structural partition, including but not limited to Z-Score standardization and physical quantity dimension consistency processing, in order to eliminate the influence of different physical field measurement scales, dimensions and amplitude ranges, and obtain calibrated multi-physics field normalized data.

[0070] S2.5 introduces time-series feature completion and anomaly detection technology to normalize multiphysics data, automatically correcting missing data and abnormal data points, ensuring the integrity and consistency of lubrication, temperature and mechanical stress signal data of each structural partition, and generating a high-quality multiphysics basic dataset.

[0071] S3: Based on multiphysics fundamental data and partition identifiers, within the global finite element model framework, a multiphysics sensitive area identification model is established for each partition, outputting the multiphysics sensitive area index and physical variable influence weights for each partition, including:

[0072] S3.1 structures the multiphysics field basic data and partition identifiers, adopts a data hierarchical mapping strategy, binds each structural partition to its associated time-series physical variables in the global finite element model, and realizes the initial construction of the partition-physical variable basic mapping table for subsequent sensitive area screening.

[0073] S3.2 Based on the constructed partition-physical variable basic mapping table, the local correlation analysis algorithm is applied to perform multidimensional statistics on the temporal distribution characteristics of multiple physical field variables such as lubrication, temperature and mechanical stress in the bearing structure partition, and obtain the correlation matrix of physical variables in the partition, which serves as the basic data for subsequent sensitivity measurement.

[0074] Based on the partition-physical variable basic mapping table, the input includes the labels of each structural partition, the temporal physical variable data of each partition (multi-physical field variables such as lubrication, temperature, and mechanical stress), and the structural partition identification information corresponding to each temporal data.

[0075] A local correlation analysis algorithm (parameters: time series data of physical variables in each partition, analysis window length, correlation criterion type) is used to perform pairwise correlation statistics on the multi-physics time series variables in each structural partition, quantify the joint change relationship of each physical field variable over time in that partition, and realize the preliminary information coupling degree extraction between variables.

[0076] Furthermore, by using time-series sliding window correlation analysis (parameters: window step size, window width), the correlation change trend of various types of physical variables within the partition at different time segments is tracked, and the deviation of correlation estimation is corrected based on the statistical sample size to ensure the robustness of variable coupling relationship under extreme or abnormal working conditions, and a preliminary correlation coefficient matrix based on time segments is obtained.

[0077] Furthermore, through a multidimensional correlation matrix aggregation algorithm (parameters: variable category, partition label, time slice range), spatial-temporal aggregation and synthesis of correlation statistics within all time-series windows are performed to form a partition-level multiphysics correlation matrix. This correlation matrix uses variable categories as rows and columns, and each element is the correlation coefficient between corresponding variable pairs (such as Pearson correlation coefficient, mutual information value, or Spearman rank correlation coefficient). The system reflects the degree of coupling between physical field variables in different partition spaces and dynamic operating conditions.

[0078] Furthermore, by using a statistical significance filtering algorithm (parameters: correlation threshold, p-value criterion, variable pair selection criteria), variable pairs with insufficient statistical sample size or no significant correlation are eliminated, and only highly correlated or highly influential regional physical variable correlation pairs are retained, providing original basic data for subsequent sensitivity measurement and principal component quantitative clustering.

[0079] Through the aforementioned local correlation analysis and correlation matrix construction, the partition-physical variable basic mapping table is transformed into a partition-level multiphysics correlation matrix, realizing the quantification of in-depth, dynamic, and spatial coupling relationships between physical variables, and providing highly reliable basic statistical data for structural partition multiphysics sensitivity screening.

[0080] For example, in a certain batch of oil-impregnated bearing global finite element models, for the rolling element region numbered A01 and the lubrication channel region numbered A02, the sampling rate of the collected multiphysics normalized time series data is 1000Hz, and the sample size is 180,000 (corresponding to a typical working condition for 3 consecutive minutes).

[0081] Pearson correlation coefficient was used as the criterion for local correlation analysis to analyze the pairwise correlations of three types of physical variables—lubrication pressure, surface temperature, and mechanical stress—within this region. The corresponding formulas are as follows:

[0082]

[0083] Where, r XY Let X be the Pearson correlation coefficient between variables X and Y. i and Y i These are the normalized physical quantity values ​​of the i-th sampling point. and , where are the sample mean and n is the number of sample points.

[0084] Using a window width of 1000 (i.e., data every 1 second) as the unit and a sliding step of 500, the correlation coefficients of the three pairs of variables (lubrication-temperature, lubrication-stress, and temperature-stress) within each window are calculated to obtain the time-correlation change sequence.

[0085] The statistical results of all windows were aggregated by mean, and weakly correlated samples with an absolute value of correlation coefficient below 0.3 were removed using p < 0.05 as the significance criterion. The multiphysics correlation matrix of region A01 was finally obtained.

[0086]

[0087] Within partition A01, the correlation coefficients between lubrication and temperature are 0.68, and between temperature and stress are 0.63, both exceeding the predetermined thresholds, thus identifying them as sensitive variable pairs. The correlation matrix of this partition serves as direct input for subsequent principal component analysis and sensitivity weight calculation, significantly improving the accuracy and engineering interpretability of physical coupling relationship identification.

[0088] S3.3 performs sensitivity quantification assessment on the correlation matrix and actual failure annotation data. It uses multivariate statistical methods such as principal component analysis (PCA) and factor weight normalization to calculate the sensitivity weight distribution of the influence of each physical variable on structural performance and outputs the multiphysics sensitivity factor matrix for each structural partition.

[0089] Joint sensitivity quantization is performed on the input partition-level multiphysics correlation matrix and actual failure annotation data to identify the degree of influence of multiphysics variables within each structural partition on structural performance.

[0090] Principal component analysis (PCA) is used (parameters: correlation matrix, cumulative contribution rate threshold, variable category) to perform eigenvalue decomposition on the multiphysics correlation matrix at the partition level, obtain each principal component vector and its corresponding eigenvalue, and realize the automatic screening of key physical factors that can explain most of the variance in the high-dimensional variable space.

[0091] Furthermore, the variance contribution of each principal component is quantitatively analyzed using a feature contribution rate calculation method (parameters: principal component eigenvalue vector, total variance). The components are then sorted in descending order to determine the top few principal components, ensuring their cumulative contribution rate meets a preset threshold (e.g., 95%). This process reduces the impact of redundant variables and highlights the physical response-sensitive dimensions.

[0092] Furthermore, the factor weight normalization algorithm (parameters: principal component loading matrix, normalization method) is applied to normalize the principal component loadings, and the absolute contribution weight of each principal component is allocated to the original physical variables, thereby quantifying the sensitivity weight of each physical variable within the structural partition.

[0093] Furthermore, based on actual failure annotation data, regression analysis or discriminant statistics (parameters: principal component score, discriminant function, failure classification label) are used to calculate the correlation between principal component score and actual failure probability or failure type. Sensitive factor residual analysis is used to screen physical variables that are highly correlated with failure and to recalibrate the sensitivity weights.

[0094] Through matrix recombination and normalization, the physical variable sensitivity data within each partition, after PCA and weight normalization, are integrated to output a multiphysics sensitivity factor matrix for each structural partition. This matrix uses physical variable categories as rows, principal component mapping indices as columns, and matrix elements as weight coefficients, comprehensively characterizing the sensitivity response of each physical variable to the partition's structural performance.

[0095] Through the principal component analysis, factor weight normalization, and failure recorrection processing described above, the results of the prior correlation statistics are effectively transformed into a multi-physics sensitive factor matrix at the structural partitioning level, enabling the quantitative identification of key physical quantities of the failure mechanism, thereby laying a data foundation for the subsequent sensitive area division and local modeling process.

[0096] For example, in the rolling body partition numbered A01, the Pearson correlation matrix is ​​input as follows:

[0097]

[0098] Principal component decomposition using PCA yielded eigenvalues ​​of λ1 = 2.1, λ2 = 0.6, and λ3 = 0.3.

[0099] The cumulative contribution rates are as follows:

[0100]

[0101] Where k is the principal component index.

[0102] Selecting the first two principal components, their combined contribution rates are: That is, more than 90%.

[0103] Principal component vectors (simplified example):

[0104] PC1: (0.61, 0.58, 0.53), PC2: (0.45, -0.82, 0.36).

[0105] The principal component loads of each variable were normalized by absolute value to obtain the sensitivity weights: lubrication 0.41, temperature 0.39, and stress 0.36. Based on the failure sample data, lubrication and temperature were determined to be the principal sensitive physical fields in zone A01.

[0106] The output multiphysics sensitivity factor matrix for partition A01 is as follows:

[0107]

[0108] Under this technical process, different working condition samples and partitioned scenarios can be adapted to the above analysis paradigm, realizing the standardization, automation and traceability of regional physical variable sensitivity identification. The output data will be used for subsequent sensitive area clustering, key sub-model setting and global coupling strategy optimization, effectively improving the ability to restore the mechanism of complex multi-physics load and the accuracy of simulation analysis.

[0109] Based on the multiphysics sensitivity factor matrix, S3.4 performs sensitive region clustering on the structural partitions in the global finite element model, uses spatial clustering algorithm and neighborhood identification mechanism to identify high-weight physical response regions, and automatically generates sensitive region indexes for each structural partition as spatial references for refined sub-model nesting.

[0110] The input conditions are: partition-level multiphysics sensitivity factor matrix, global finite element model structural partition information, and spatial distribution data of physical variables in the structural partition.

[0111] A spatial clustering algorithm (parameters: sensitivity factor threshold, neighborhood radius, variable weight vector) is used to perform high-dimensional feature clustering on each structural partition in the global finite element model, thereby realizing the automatic aggregation of multi-physics sensitivity factors in space and identifying a set of nodes or units with similar physical response characteristics.

[0112] Furthermore, through a high-weight physical response region identification mechanism (parameters: lower limit of sensitivity weight, spatial adjacency constraint), based on the weight distribution in the sensitivity factor matrix, a weighted screening is performed on the nodes or units in the spatial clustering results to form a set of spatial candidate regions with weights significantly higher than the average level.

[0113] Furthermore, a neighborhood search and spatial connectivity detection algorithm (parameters: node adjacency matrix, connectivity threshold) is adopted to find sub-regions with continuous boundaries and coordinated responses within the candidate region set, and to exclude scattered nodes that have no physical coupling significance, so as to ensure that sensitive regions have high physical correlation and structural coherence.

[0114] Furthermore, based on the spatial optimization function (parameters: region volume, physical variable gradient distribution, model boundary conditions), the distribution boundary of sensitive regions is optimized, and the position and range of cluster partitions are adjusted so that the clustering results of sensitive regions can cover the highly sensitive regions to the greatest extent, while avoiding redundancy and overlap, and improving the spatial utilization efficiency of subsequent sub-model nesting.

[0115] By using a numbering and indexing algorithm, the sensitive areas within each structural partition are encoded with unique identifiers to generate a structural partition-sensitive area index table, thereby enabling the automatic generation and location tracking of the sensitive area index.

[0116] Through the above spatial clustering and neighborhood calibration algorithms, the numerical results of the multi-physics sensitivity weights of the partition are transformed into spatial sensitive area index data, providing a strict spatial reference for subsequent refined sub-model nesting, and realizing the high degree of adaptability and accurate positioning of sensitive area modeling objects.

[0117] For example, in the A01 rolling element partition of the global finite element model of the oil-impregnated bearing, based on the obtained sensitivity factor weights (lubrication weight 0.41, temperature weight 0.39, stress weight 0.36), the spatial clustering algorithm based on DBSCAN is used, with the minimum number of neighborhood elements set to 12 and the sensitivity weight threshold set to 0.3, to perform weighted aggregation of the three types of physical field response values ​​of each grid node.

[0118] For the high-weight region, further node connectivity analysis identified two main spatially connected sub-regions: the first region covers the outer ring-lubricating film interface of the rolling element, and the second region is located in the relative slip zone between the rolling element and the cage. After spatial boundary optimization and exclusion of isolated low-response nodes, the first sensitive region accounts for 18.7% of the volume of partition A01, and the second region accounts for 6.2%.

[0119] Assign a unique index to each sub-region (e.g., A01-R1, A01-R2), and generate a region index table A01:[A01-R1,A01-R2].

[0120] The verification process shows that the clustering and calibration results can accurately cover the region with the maximum gradient of oil film pressure and the region with concentrated temperature rise, providing a complete spatial basis for subsequent local high-precision modeling. Moreover, the coupling correlation between oil film thickness and contact stress in the sensitive sub-region is improved to 0.76 (higher than the partition average of 0.51), which greatly improves the accuracy of physical field identification and dynamic response analysis of the model.

[0121] S3.5 integrates the sensitive region index of the partition with its corresponding multiphysics sensitivity weights, and outputs a structural partition-sensitive region-weight mapping table to achieve traceable physical variable influence, locatable position, and quantitative response. This provides accurate basic data for the selection setting and parameter initialization of subsequent local high-precision multiphysics coupled micro-element sub-models.

[0122] Step S4: For the identified multiphysics sensitive regions, establish a local high-precision multiphysics coupled micro-element model, including modeling the lubrication fluid dynamics-structure thermal coupling behavior, and associate the micro-element model parameters with the sensitive region index from the previous step. Specifically, this includes:

[0123] S4.1 uses a multiphysics sensitive region index to filter the corresponding local structural geometric parameters, material parameters, and working boundary conditions to generate a multiphysics sensitive region micro-element analysis object, which serves as the data input for high-precision local modeling.

[0124] The multiphysics sensitive region index, local structural geometric parameter set, material parameter set, and working condition boundary condition parameters are used as data inputs.

[0125] A sensitive region index filtering algorithm (parameters: structural partition-sensitive region-weight mapping table, high weight threshold, spatial index number) is adopted to realize the automatic spatial positioning of the multi-physics sensitive region calibrated in the global model of the target oil-impregnated bearing.

[0126] Furthermore, by using a local structural geometric parameter extraction method (parameters: 3D CAD model, node coordinate set, sensitive area boundary), the precise geometric dimensions, spatial positions, surface division contours, and other data of each node and unit within the sensitive area are exported in batches, and a corresponding subset of geometric parameters is constructed to ensure that the local model can be modeled in detail at high resolution.

[0127] Furthermore, a local material property aggregation algorithm (parameters: material identifier, physical property database, sensitive area unit list) is adopted to classify and retrieve the elastic modulus, Poisson's ratio, density, coefficient of thermal expansion, thermal conductivity and oil film physical property parameters of various materials such as matrix, lubrication pair, and cage involved in the sensitive area, forming a multiphysics micro-element material parameter table.

[0128] Furthermore, by employing a dynamic working condition mapping algorithm (parameters: working condition partition mapping table, corresponding working condition label, stress-load-temperature time series data), the historical working condition time series boundary conditions of the sensitive area partition are matched, including the integration of multiple time series boundary parameters such as dynamic load, impact load magnitude and direction, lubrication oil supply rate, heat flux density, and ambient temperature.

[0129] Furthermore, a multiphysics micro-element object generation algorithm (parameters: sensitive region index, local geometric parameters, material parameters, and working condition boundary conditions) is adopted to encapsulate all the above local input parameters into multiphysics sensitive region micro-element analysis objects for high-precision local modeling. Each micro-element object data has a unique spatial, physical, and working condition attribute identifier, which facilitates subsequent modeling and adaptive initialization of coupling interfaces.

[0130] Through the above-mentioned multi-algorithm chain processing, the sensitive areas identified in the global finite element model are intelligently and efficiently transformed into structured local high-precision multiphysics micro-element analysis objects, realizing the preparation of pre-parameters for refined modeling of sensitive areas.

[0131] For example, in the rolling element-lubricating film interface sensitive area (A01-R1) of partition A01, with a sensitive area volume of 1.2 mm... 3 As the test object, 1132 nodes and 786 units were automatically located using a spatial index filtering algorithm;

[0132] By extracting local geometric parameters, the node coordinate ranges [x1, x2], [y1, y2], and [z1, z2] are obtained, along with the average surface roughness R. a It is 0.12 μm;

[0133] Local material parameters were retrieved for steel (elastic modulus 209 GPa, Poisson's ratio 0.28) and lubricating oil (dynamic viscosity 0.024 Pa·s, thermal conductivity 0.13 W / m·K);

[0134] Using a working condition mapping algorithm, the load timing range is 25N~240N, the maximum pulse impact load is 320N, the oil temperature timing range is 22℃~97℃, and the lubrication flow rate is 5.6imes10. -4 m^3 / s;

[0135] Through the multiphysics micro-element object generation algorithm, the micro-element object of this sensitive region has included a node-element list, a material-property parameter table, and a working condition boundary time series parameter set. Finally, it outputs a complete micro-element object data package for S4.2 mesh refinement and S4.3 multiphysics field coupling solver initialization.

[0136] The standardized output of this object significantly improves the accuracy and efficiency of local modeling under complex loads and multiphysics scenarios.

[0137] S4.2 employs a multi-scale finite element meshing algorithm to refine the finite element mesh structure of the micro-element analysis object in the multi-physics sensitive region, so as to meet the local high-precision requirements and enhance the spatial resolution of node information transmission, thereby obtaining a refined finite element mesh model.

[0138] The input for multiphysics sensitive region micro-element analysis includes screened local geometric parameters, material parameters, and dynamic working condition boundary conditions, and has a unique spatial and physical identifier.

[0139] A multi-scale finite element meshing algorithm (parameters: target sensitive area node / element list, initial mesh size, refinement scale interval, accuracy grading strategy) is adopted to refine the spatial region defined by the micro-element analysis object, thereby gradually improving the spatial dispersion within the sensitive area.

[0140] Furthermore, by using a region-adaptive mesh generation algorithm (parameters: local physical quantity change rate threshold, maximum element size limit, and boundary layer automatic refinement factor), the mesh density of key areas such as oil film interfaces, high stress concentration points, and temperature rise abrupt change zones is automatically optimized based on the physical response gradient distribution, thereby enhancing the node distribution density in these areas and improving spatial resolution.

[0141] Furthermore, to address the spatial coupling requirements of different physical fields, a heterogeneous multi-physics coupling mesh reconstruction technique is adopted (parameters: discretization format of each physical field, coupling node pair mapping table). This technique establishes node mapping relationships and physical quantity projection functions between the fluid, structural, and thermal physical field sub-regions, ensuring that the transmission values ​​of physical quantities between the multi-physics field sub-models are consistent and the errors are controllable.

[0142] Furthermore, by refining the mesh quality optimization algorithm (parameters: element aspect ratio, acute angle threshold, boundary fitting accuracy standard), the generated local mesh structure is reconstructed, nodes are redistributed, and boundary contours are refitted, effectively avoiding the numerical instability risks of isolated weak elements and high deformation zones.

[0143] Furthermore, a multi-level data encapsulation scheme for the mesh structure (parameters: level number, node-element correspondence table, coupling interface mark) is adopted to archive the refined finite element mesh structure in the form of layered objects, which facilitates subsequent high-precision multiphysics coupling solution and dynamic parameter tuning, and finally obtains a refined finite element mesh model that accurately corresponds to the sensitive region index.

[0144] Through the above-mentioned multi-level and multi-algorithm coordinated refinement process, the micro-element objects in the sensitive area of ​​the multiphysics field are structurally transformed into locally refined finite element meshes with high spatial resolution and consistent physical quantity coupling. This provides standardized modeling units for subsequent high-precision coupled simulation analysis of multiphysics fields and dynamic iteration of physical field parameters, thereby significantly improving the analytical capability of physical response and the accuracy of numerical prediction in the sensitive area.

[0145] For example, in the sensitive region of a typical oil-impregnated bearing A01-R1, the initial micro-element object has 1132 nodes, 786 elements, and a region volume of 1.2 mm. 3 A multi-scale finite element meshing algorithm was adopted, with the maximum allowable element side length set at 0.05 mm, the minimum side length at 0.01 mm, and the threshold for the rate of change of regional physical quantities at 5%, to refine the mesh at the lubricating oil film interface and the high gradient stress zone.

[0146] As a result, the total number of elements increased to 2684, the number of nodes increased to 3497, and the minimum element size at the local oil film interface reached 0.009 mm. Through heterogeneous multiphysics coupled mesh reconstruction, 364 sets of key nodes were mapped between the three subfields of lubrication, structure, and heat.

[0147] A mesh quality optimization algorithm is adopted, the aspect ratio of the cells is controlled between 1 and 4, all acute angles are greater than 30°, and the boundary profile error is less than 0.5%.

[0148] Ultimately, the refined finite element mesh model effectively improved the node density and spatial resolution. The local relative error of the numerical prediction of oil film thickness and contact stress in sensitive areas was less than 2.1%, supporting the next step of multiphysics coupling solution to achieve higher spatiotemporal accuracy and engineering interpretability.

[0149] S4.3 utilizes a fluid-structure-thermal multiphysics coupled solver to apply fluid dynamics, heat conduction, and structural mechanics boundary conditions to the lubricating oil film sub-region, heat source point, and bearing surface in the refined finite element mesh model, thereby obtaining the joint response parameters of the lubricating fluid dynamics-structural thermal multiphysics.

[0150] The analysis takes the micro-element object of the multi-physics sensitive region and its refined finite element mesh model as input. The micro-element object includes the extracted geometric parameters, material parameters and multi-time-series boundary conditions. The refined mesh contains high spatial resolution nodes and element attributes.

[0151] A fluid-structure-thermal multiphysics coupled solver (parameters: fluid subfield coupling scheme, structural subfield finite element characteristics, heat conduction solution module) is used to define the physical field boundary types for each mesh element and node in the sensitive region, including fluid dynamic boundary (inlet pressure, viscous oil film shear velocity), heat source boundary conditions (local heat flux density, adiabatic / isothermal surface), and structural mechanical boundary conditions (load transfer surface, constrained surface, nodal displacement constraint), to realize the multiphysics boundary configuration of lubricating oil film sub-region, heat source point and bearing surface.

[0152] Furthermore, based on a partitioned iterative algorithm (parameters: iteration step size, subfield coupling strength, convergence criterion) using the fluid-structure / thermal three-field coupled solution, the fundamental physical equations governing lubrication flow, heat transfer, and structural deformation are solved simultaneously, progressing according to the time step. In the fluid dynamics sub-region, the incompressible Navier-Stokes equations are used to describe the flow behavior within the lubricating oil film:

[0153]

[0154] Where ρ is the density of the lubricating oil, u is the velocity vector, p is the pressure, μ is the viscosity, and f is the volume force.

[0155] In the heat conduction sub-region, the transient heat conduction equation is used to describe the local temperature rise at the heat source point and on the wall:

[0156]

[0157] Among them, c p λ is the specific heat capacity, T is the temperature field, λ is the thermal conductivity, and Q is the heat source per unit volume.

[0158] In the structural mechanics sub-region, the force and displacement responses of the nodes are solved using the finite element method, employing equilibrium equations:

[0159] Ku = F ext +F int +F th

[0160] Where K is the global stiffness matrix, u is the nodal displacement vector, and F ext For external load, F int For the oil film-structure contact internal force, F th This refers to the thermal load caused by thermal expansion.

[0161] Furthermore, through the subfield coupling-mapping-synchronization mechanism (parameters: cascaded boundary mapping table, node physical quantity transfer factor, convergence synchronization tolerance), at each time integration step, the internal pressure distribution of the lubricating oil film is mapped to the structural nodes of the bearing surface, and the temperature rise of the heat source point is synchronized to provide feedback on the fluid viscosity and structural thermal expansion, thereby realizing the dynamic transfer of physical properties and boundary conditions among the three fields of fluid, heat, and structure.

[0162] Furthermore, a coupled field convergence judgment and error control algorithm (parameters: maximum allowable residual, coupling criterion type) is adopted to judge the convergence of the main variables of pressure, temperature and displacement in each iteration. If the residuals of all main variables are lower than the preset threshold, the multi-physics joint response parameters of the current step are output; otherwise, local re-relaxation correction is automatically performed until convergence.

[0163] By using a fluid-structure-thermal multiphysics coupled solver and a multi-step dynamic boundary transfer algorithm, the joint response parameters of oil film pressure, flow velocity field, temperature distribution, nodal displacement, and stress response of the lubricating oil film sub-region, heat source point, and bearing surface with respect to time are obtained within a refined finite element sensitive region mesh, thus achieving high-fidelity multiphysics time-series parameter output.

[0164] For example, in the micro-element model of the sensitive region numbered A01-R1, the fluid subfield adopts the incompressible Navier-Stokes equation, and the oil film density ρ = 860 kg / m³. 3 The dynamic viscosity μ = 0.024 Pa·s; the thermal field adopts the transient heat conduction equation, with a thermal conductivity λ = 0.13 W / m·K and an initial temperature T0 = 40℃; the structural subfield has a steel stiffness modulus E = 209 GPa. The operating condition input is a dynamic load range of 25 N to 240 N, with an oil film inlet velocity u. in =0.22m / s, periodic endogenous heat rate at the heat source point Q = 19.4kW / m 3 The time step width is set to 0.001s, the subfield coupling strength parameter to 0.8, and the convergence tolerance to 1×10⁻⁶. -3In the iterative solution of the fluid-structure thermal field, at each iteration, the oil film pressure field of the current time step is applied to the structural bearing surface mesh through node mapping. The temperature response of the structural node deformation is fed back to the thermodynamic field to adjust the temperature and local viscosity of the oil film region. Under multi-step convergence criteria, all residuals are below 1×10⁻⁶. -4 Subsequently, the pressure profile, temperature distribution, and nodal deformation data for each step were stably output. Ultimately, within a 0.3-second loading cycle, 210 sets of high-frequency oil film pressure-structural stress-local temperature multi-field correlation data were acquired. The extreme dynamic variation range of the oil film pressure was 1.9–3.3 MPa, the maximum temperature rise zone reached 91.2℃, and the maximum equivalent stress at the contact master node was 143 MPa. The above high-fidelity joint response parameter output provides a solid data foundation for subsequent failure criterion extraction and model dynamic optimization.

[0165] Based on the aforementioned multi-physics field joint response parameters, S4.4 executes a dynamic prediction algorithm for surface contact stress and oil film thickness, outputting high-resolution local parameters such as time-series oil film pressure distribution, temperature rise field, and contact area stress response sequence, forming a local high-precision multi-physics field response set.

[0166] S4.5 establishes a correlation data mapping rule between the intermediate parameters of the local high-precision multiphysics response set and the original global finite element model through the sensitive region index, realizing the structural synchronous correlation between the parameters of the local micro-element sub-model and the sensitive region index, and providing a standardized interface for subsequent multi-scale multiphysics coupling and high-speed information exchange.

[0167] S5: Inputting the working condition data sample and the local multiphysics coupled micro-element sub-model into the global finite element model, and realizing high-speed communication between global and local multiphysics information through a multi-scale mapping method, and outputting multi-scale multiphysics interactive data specifically includes:

[0168] S5.1 performs semantic label mapping on the obtained working condition data samples, takes working condition parameters such as dynamic load, impact load and asymmetric load as input, and performs data adaptation processing on each structural partition through the working condition partition mapping table to generate partition working condition parameter set, providing multi-dimensional time-series working condition driving factors for subsequent finite element modeling.

[0169] S5.2 is based on the partitioned working condition parameter set and adopts a multi-physics coupling mapping algorithm. It integrates the modeling parameters, initial conditions and sensitive region index of the local multi-physics coupled micro-element sub-model as parameter factors with the node attribute data of the global finite element model to generate partitioned multi-scale multi-physics initial configuration, thereby realizing parameter synchronization and state initialization between partitions.

[0170] S5.3 utilizes partitioned multi-scale multi-physics initial configuration to construct a multi-physics mapping interface for the global finite element model. Through multi-scale mesh coupling techniques such as matrix mapping and node interpolation, it achieves high-speed communication of physical variables between the local multi-physics coupled micro-element sub-model and the global finite element model, completing the model coupling initialization under working condition drive.

[0171] S5.4 is based on the model coupling initialization state and adopts a state timing synchronization mechanism to drive the global finite element model to perform multi-physics joint simulation iteration under dynamic load timing. It transmits the time-condition parameters to each local multi-physics coupled micro-element sub-model and receives the high-precision local physical field response returned by the local model, realizing bidirectional feedback of global-local multi-physics dynamic data.

[0172] S5.5 integrates the multi-physics information of the structural partitions through a partition reconstruction algorithm on the real-time returned data of the global finite element model and the local multi-physics coupled micro-element sub-model. It extracts a multi-scale multi-physics interactive dataset containing multi-dimensional numerical features such as fluid pressure, oil film thickness, nodal temperature, and contact stress, laying the data foundation for subsequent machine learning inversion and durability evaluation processes.

[0173] S6: Based on multi-scale multi-physics interaction data, a machine learning model is used to predict and invert the local and global physical responses driven by time-series load parameters, calibrate the results of traditional physical modeling, and generate multi-physics coupling calibration factors, specifically including:

[0174] S6.1 performs feature engineering on multi-scale multi-physics interaction data to extract time-series feature parameters that are highly correlated with local physical field interactions and global dynamic load responses, generating a physical field interaction feature set.

[0175] S6.2 Based on the physical field interaction feature set, a preliminary machine learning mapping model is established for different structural partitions and sensitive regions. This model includes a supervised learning model or a graph neural network model, which is used to characterize the high-dimensional nonlinear interaction relationship between local and global physical responses and output local-global response prediction parameters.

[0176] S6.3 compares historical experimental data, numerical simulation results, and local-global response prediction parameters, and uses transfer learning or data inversion algorithms to automatically identify systematic deviations between the physical model and actual observations, generating calibration residuals for the physical response model.

[0177] S6.4 uses a physical response model to calibrate the residuals. By optimizing traditional physical modeling parameters (such as material constitutive parameters, lubrication coefficient, thermal conductivity, etc.), the machine learning mapping model is iteratively trained until the error between the prediction result and the observed data meets the predetermined convergence criterion, and the calibrated multiphysics coupling mapping factor is obtained.

[0178] S6.5 outputs the calibrated multiphysics coupling mapping factor and the calibrated local-global physical response prediction model, which are used to dynamically adjust the parameters of the subsequent global finite element model and local high-precision sub-model, supporting continuous adaptive optimization of high-fidelity co-simulation of multiphysics and dynamic loads.

[0179] S7: Utilizing multiphysics coupling calibration factors to dynamically adjust the parameters of the global finite element model and the local high-precision micro-element sub-model, achieving highly adaptable modeling for extreme and atypical working conditions, specifically includes:

[0180] S7.1 performs categorized analysis on the multiphysics coupling calibration factors generated by the machine learning model. Based on the structural unit types and node physical attribute distribution in the global finite element model, it executes a parameter mapping strategy to generate a global finite element model calibration parameter set, so as to realize the correspondence between the multiphysics coupling calibration factors and the dynamic parameters of the global finite element model.

[0181] S7.2, based on the global finite element model calibration parameter set, performs batch dynamic assignment and real-time adjustment of the physical property parameters of the global finite element model (including nodal stiffness matrix, thermal conductivity, friction contact parameters, etc.), and outputs the dynamically adjusted global finite element model parameter array to ensure the high adaptability of the global finite element model to the physical field cooperative response under extreme and atypical dynamic conditions.

[0182] S7.3 performs regional positioning of multi-physics field coupled calibration factors. Based on the multi-physics field sensitive region index in the local high-precision micro-element sub-model, it extracts the parameter vector of the corresponding sub-factor and generates a dynamic calibration parameter set exclusive to the micro-element sub-model to achieve hierarchical mapping of global and local calibration parameters.

[0183] S7.4 applies the dynamic calibration parameter set specific to the micro-element sub-model to local physical fields such as lubricating fluid, structural stress, and heat conduction. Through micro-element modeling algorithms such as nested numerical solution and real-time adjustment of physical field boundary conditions, it performs dynamic parameter correction on the local high-precision multi-physics coupled micro-element sub-model to obtain the adjusted micro-element sub-model input parameter set.

[0184] S7.5 summarizes and adjusts the global finite element model parameter array and the micro-element sub-model input parameter group, and uses a multi-scale multi-physics mapping algorithm to achieve synchronous communication between the two layers of information. The feedback information of dynamic parameter adjustment is fed back to the machine learning model that generates multi-physics coupling calibration factors in real time, realizing closed-loop optimization control of adaptive modeling parameters under complex working conditions, and outputting a highly adaptable multi-physics model set for subsequent simulation analysis.

[0185] S8: Extracting time-series features from the multiphysics co-simulation results after dynamic parameter adjustment to obtain multi-dimensional parameter sequences related to typical failure mechanisms such as oil film thickness distribution, fluid pressure, temperature rise, and contact stress. Specifically, this includes:

[0186] S8.1 performs multi-dimensional physical field data extraction in different regions based on the multi-physics joint simulation results after dynamic parameter adjustment. Relying on the global finite element model region index, oil film thickness, fluid pressure, temperature rise and contact stress are used as input variables. Based on the partition number, spatial mapping is performed to output partition-level multi-physics time series raw data.

[0187] S8.2 employs a physical field adaptive preprocessing algorithm for filtering and normalization of the original multi-physics time series data at the partition level, in order to remove high-frequency noise and scale drift driven by operating conditions, thereby generating highly consistent, multi-physics normalized time series data.

[0188] S8.3 takes multi-physics normalized time series data as input and uses a sliding window-based dynamic feature extraction method to gradually segment the oil film thickness distribution, fluid pressure change, temperature rise curve and contact stress sequence within a continuous time period. This enables the regional extraction and discrimination of sensitive feature parameters of typical failure modes (such as minimum oil film thickness, maximum instantaneous fluid pressure, extreme temperature rise point and local stress peak).

[0189] S8.4 uses time-series statistical criteria and frequency domain analysis algorithms to further process the response modes of extreme operating conditions for sensitive characteristic parameters, and extracts operating condition characteristic sequence indicators that reflect the failure mechanism, including time-series structured characteristics such as periodic abrupt changes in oil film thickness, abnormal fluctuations in fluid pressure, continuous accumulation rate of temperature rise, and time-varying distribution of contact stress.

[0190] S8.5 combines temporal structured features with the original partition number and, based on the spatial heat map reconstruction algorithm of partition-level multi-physics sensitive parameters, outputs multi-dimensional parameter sequences related to the global typical failure mechanism of bearings, such as oil film thickness distribution, fluid pressure flow field, local temperature rise distribution, and contact stress cloud map, forming a spatial-temporal multi-physics feature database for durability criterion evaluation and subsequent adaptive iterative optimization of the model.

[0191] S9: Determine whether the extracted multidimensional parameter sequence meets the preset durability and failure criteria. If not, automatically optimize the modeling granularity and coupling ratio of the multiphysics sensitive region based on feedback information to form an adaptive model iteration, specifically including:

[0192] S9.1 performs a criterion comparison process on the multi-dimensional parameter sequences such as oil film thickness distribution, fluid pressure, temperature rise and contact stress in the multi-physics field joint simulation results after dynamic parameter adjustment, based on the pre-set durability criterion and failure criterion template parameters, to obtain the criterion achievement identification result.

[0193] S9.2 Based on the obtained criteria and identification results, call the feature deviation analysis algorithm to quantitatively calculate the technical deviation between the non-compliant multidimensional parameter sequence and the preset criteria, and output the main influencing factors and corresponding suggestions for changing the multiphysics sensitive area.

[0194] S9.3 Based on the multiphysics sensitive area change suggestions, execute the sensitive area modeling granularity optimization algorithm to dynamically refine or coarsen the spatial resolution and physical variable sampling density of the original multiphysics sensitive area, and form optimized modeling granularity distribution data.

[0195] S9.4 takes the optimized modeling granularity distribution data as input, uses the coupling weight adaptive adjustment algorithm to dynamically adjust the weights of the multi-physics coupling parameters, and outputs the updated regional coupling weight distribution matrix.

[0196] S9.5 reconfigures the information interaction parameters between the multiphysics coupled micro-element sub-model and the global finite element model based on the updated regional coupling weight distribution matrix, realizes the adaptive simulation configuration optimization of the model, generates a new round of multiphysics joint simulation input conditions, and realizes the adaptive iterative closed loop of the model.

[0197] S10: Outputs adaptively optimized high-fidelity multiphysics-dynamic load joint simulation analysis results, used for predicting the durability and failure mechanism of oil-impregnated bearings under complex dynamic conditions and for authoritative load-bearing capacity assessment, specifically including:

[0198] S10.1 normalizes and structures the multidimensional time-series parameter sequence output by the multiphysics co-simulation after dynamic parameter adjustment, based on the feature mapping algorithm, to obtain a unified format of oil film thickness distribution parameter sequence, fluid pressure parameter sequence, temperature rise parameter sequence and contact stress parameter sequence, as the basic input for subsequent durability and failure mechanism evaluation.

[0199] S10.2 Based on the structured multidimensional time-series parameter sequence, the working condition adaptive failure criterion algorithm is used to perform multi-criterion aggregation evaluation on oil film thickness distribution parameters, fluid pressure parameters, temperature rise parameters and contact stress parameters, so as to achieve preliminary classification and prediction of the durability and failure risk of oil-impregnated bearings under combined working conditions.

[0200] S10.3 Based on the preliminary classification and prediction of failure risk information, high-fidelity multiphysics field-dynamic load inversion technology is used to conduct engineering source analysis on local extreme response characteristics, further refine the fault failure mechanism diagnosis output, and obtain regional failure causes and evolution paths.

[0201] Based on the failure mechanism diagnosis results, S10.4 uses a model adaptive evaluation algorithm to perform secondary parameter self-optimization on simulation modeling granularity, sensitive area coupling parameters, etc., to generate the final optimal multiphysics modeling parameter set and a high-fidelity simulation database of global dynamic load response.

[0202] S10.5 takes the optimal multiphysics modeling parameter set and high-fidelity simulation database as input, and uses an authoritative load-bearing capacity assessment model to quantitatively output the load-bearing capacity boundary, durability life, and ultimate failure mode of oil-impregnated bearings under various complex dynamic working conditions in multiple dimensions, forming a deliverable authoritative load-bearing capacity assessment report.

[0203] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this application.

Claims

1. A method for evaluating the load-carrying capacity of oil-impregnated bearings based on virtual simulation, characterized in that, include: S1. Obtain working condition data samples of oil-impregnated bearings under typical operating conditions under load conditions, and identify the bearing structure partitions to establish a working condition partition mapping table. S2. For different structural partitions in the working condition partition mapping table, collect multiphysics time series experimental data, and perform noise reduction and standardization processing to obtain multiphysics basic data. S3. Based on multiphysics field basic data and partition identifiers, under the global finite element model framework, establish a multiphysics field sensitive area identification model for each partition, and output the multiphysics field sensitive area index and physical variable influence weight of each partition. S4. For the identified multiphysics sensitive regions, establish local multiphysics coupled micro-element sub-models, and associate the parameters of the local multiphysics coupled micro-element sub-models with the multiphysics sensitive region indexes through data mapping rules. Specifically, this includes: Based on the multiphysics sensitive region index, the corresponding local structural geometric parameters, material parameters and working condition boundary conditions are selected to generate multiphysics sensitive region micro-element analysis objects. For the micro-element analysis object in the multi-physics sensitive region, a multi-scale finite element meshing algorithm is used to refine its finite element mesh structure and obtain a refined finite element mesh model. Using a fluid-structure-thermal multiphysics coupled solver, fluid dynamics, heat conduction and structural mechanics boundary conditions are applied to the lubricating oil film sub-region, heat source point and bearing surface in the refined finite element mesh model to obtain the joint response parameters of lubricating fluid dynamics-structural thermal multiphysics. Based on the multi-physics joint response parameters, a dynamic prediction algorithm for surface contact stress and oil film thickness is executed to output high-resolution local parameters and obtain a local high-precision multi-physics response set. The intermediate parameters of the local high-precision multiphysics response set are correlated with the original global finite element model through the sensitive region index, and a correlation data mapping rule is established for the structural synchronous correlation between the parameters of the local micro-element sub-model and the sensitive region index. S5. Input the working condition data sample and the local multiphysics coupled micro-element sub-model into the global finite element model, and output multi-scale multiphysics interactive data. S6. Based on multi-scale multi-physics field interactive data, predict and invert the local and global physical responses driven by time-series load parameters, and generate multi-physics field coupling calibration factors. S7. Dynamic parameter adjustment of the global finite element model and the local high-precision micro-element sub-model is performed using the multiphysics coupling calibration factor. S8. Extract time-series features from the multiphysics joint simulation results after dynamic parameter adjustment to obtain multidimensional parameter sequences related to typical failure mechanisms.

2. The method for evaluating the load-bearing capacity of an oil-impregnated bearing based on virtual simulation according to claim 1, characterized in that, The operating condition data in S1 includes labels for dynamic load, impact load, and asymmetric load. The step of identifying bearing structure partitions to establish an operating condition partition mapping table specifically includes: Dynamic load tests were conducted on the target oil-impregnated bearing under typical operating conditions to obtain raw operating condition data covering three types of working conditions: dynamic load, impact load, and asymmetric load, forming a load type mapping dataset. Based on the original working condition data samples, an event-driven annotation algorithm is used to assign load condition labels to each time series sampling point to generate a working condition data time series label set. A partitioning algorithm is applied to the 3D CAD structural model of the oil-impregnated bearing to subdivide the bearing as a whole into functional regions and obtain the structural partition index of each region. The time series label set of working condition data and the structural partition index are fused into high-dimensional data. A multi-label mapping strategy is used to generate an integrated working condition-structure mapping table for each structural partition under various load conditions. Based on the integrated working condition-structure mapping table, data consistency verification and redundancy filtering are performed on typical structural partition samples under various operating conditions to obtain the working condition partition mapping table.

3. The method for evaluating the load-bearing capacity of oil-impregnated bearings based on virtual simulation according to claim 1, characterized in that, The multiphysics sensitive region index and physical variable influence weights of each partition output in S3 include: The multiphysics field basic data and partition identifiers are structured and organized, and a data hierarchical mapping strategy is adopted to bind each structural partition and its associated time series physical variables in the global finite element model. Based on the established partition-physical variable basic mapping table, the local correlation analysis algorithm is applied to perform multidimensional statistics on the time-series distribution characteristics of multiple physical field variables within the bearing structure partition, and obtain the correlation matrix of physical variables within the partition. Sensitivity quantification assessment is performed on the correlation matrix and actual failure annotation data. The sensitivity weight distribution of the influence of each physical variable on the structural performance is calculated, and the multi-physics sensitivity factor matrix of each structural partition is output. Based on the multiphysics field sensitivity factor matrix, the structural partitions in the global finite element model are divided into sensitive regions by clustering. The spatial clustering algorithm and neighborhood identification mechanism are used to identify the high-weight physical response regions and automatically generate the sensitive region index of each structural partition. The partition-sensitive region index and its corresponding multiphysics sensitivity weight are integrated together.

4. The method for evaluating the load-bearing capacity of oil-impregnated bearings based on virtual simulation according to claim 1, characterized in that, The S5 output multi-scale multiphysics interaction data specifically includes: Semantic label mapping is performed on the obtained working condition data samples. The working condition parameters are used as input, and the data adaptation processing of each structural partition is performed through the working condition partition mapping table to generate partition working condition parameter sets. Based on the partitioned working condition parameter set, a multi-physics coupling mapping algorithm is adopted to integrate the modeling parameters, initial conditions and sensitive region index of the local multi-physics coupled micro-element sub-model as parameter factors with the node attribute data of the global finite element model to generate a partitioned multi-scale multi-physics initial configuration. By utilizing the partition-level multi-scale multiphysics initial configuration, a multiphysics mapping interface for the global finite element model is constructed. Through multi-scale mesh coupling technology, the model coupling initialization under the condition-driven approach is completed. Based on the model coupling initialization state, a state timing synchronization mechanism is adopted to drive the global finite element model to perform multi-physics joint simulation iteration under dynamic load timing, transmit the time-condition parameters to each local multi-physics coupled micro-element sub-model, and receive the high-precision local physical field response returned by the local model. The real-time return data of the global finite element model and the local multiphysics coupled micro-element sub-model are integrated with the structural partition multiphysics information through a partition reconstruction algorithm to extract a multi-scale multiphysics interactive dataset with multi-dimensional numerical features.

5. The method for evaluating the load-bearing capacity of oil-impregnated bearings based on virtual simulation according to claim 1, characterized in that, The S6 generation of the multiphysics coupling calibration factor specifically includes: Feature engineering is performed on multi-scale multi-physics interaction data to extract time-series feature parameters that are highly correlated with local physical field interactions and global dynamic load responses, thereby generating a physical field interaction feature set. Based on the physical field interaction feature set, a preliminary machine learning mapping model is established for different structural partitions and sensitive regions, including supervised learning models or graph neural network models, to output response prediction parameters. By comparing historical experimental data, numerical simulation results and response prediction parameters, the systematic deviation between the physical model and actual observations is automatically identified, and calibration residuals of the physical response model are generated. The residuals are calibrated using a physical response model. By optimizing the parameters of traditional physical modeling, the machine learning mapping model is iteratively trained until the error between the prediction results and the observed data meets the predetermined convergence criterion, thus obtaining the calibrated multiphysics coupling mapping factor. Output the calibrated multiphysics coupling mapping factor and the calibrated physical response prediction model.

6. The method for evaluating the load-bearing capacity of an oil-impregnated bearing based on virtual simulation according to claim 1, characterized in that, The S7 achieves highly adaptive modeling for extreme and atypical operating conditions, specifically including: A categorized analysis was performed on the multiphysics coupling calibration factors generated by the machine learning model. Based on the structural unit type and node physical attribute distribution in the global finite element model, a parameter mapping strategy was executed to generate a global finite element model calibration parameter set. Based on the global finite element model calibration parameter set, the physical property parameters of the global finite element model are dynamically assigned and adjusted in batches in real time, and the dynamically adjusted global finite element model parameter array is output. The multi-physics coupling calibration factor is located in a region. Based on the multi-physics sensitive region index in the local high-precision micro-element sub-model, the parameter vector of the corresponding sub-factor is extracted to generate a dynamic calibration parameter set specific to the micro-element sub-model. The dynamic calibration parameter set specific to the micro-element sub-model is applied to the local physical field. Through the micro-element modeling algorithm, the local high-precision multi-physics coupled micro-element sub-model is dynamically corrected to obtain the adjusted micro-element sub-model input parameter set. The adjusted global finite element model parameter array and the micro-element sub-model input parameter group are summarized and fed back in real time to the machine learning model that generates multiphysics coupling calibration factors, and outputs a highly adaptable multiphysics model set for subsequent simulation analysis.

7. The method for evaluating the load-bearing capacity of an oil-impregnated bearing based on virtual simulation according to claim 1, characterized in that, The S8 process for obtaining the multidimensional parameter sequence specifically includes: The multi-physics joint simulation results after dynamic parameter adjustment are subjected to regional extraction of multi-dimensional physical field data. Based on the global finite element model region index, oil film thickness, fluid pressure, temperature rise and contact stress are used as input variables. Based on the partition number, spatial mapping is performed to output partition-level multi-physics time series raw data. For the original multiphysics time series data at the partition level, high-frequency noise and scale drift driven by operating conditions are removed to generate multiphysics normalized time series data. Using multi-physics normalized time series data as input, a sliding window-based dynamic feature extraction method is used to progressively segment the oil film thickness distribution, fluid pressure changes, temperature rise curves, and contact stress sequence within a continuous time period, thereby obtaining sensitive characteristic parameters of typical failure modes. For sensitive characteristic parameters, time-series statistical criteria and frequency domain analysis algorithms are used to further process the response mode of extreme working conditions, extract the working condition characteristic sequence index that reflects the failure mechanism, and obtain time-series structured characteristics. By combining temporal structured features with the original partition numbers, and based on the spatial heatmap reconstruction algorithm of partition-level multiphysics sensitive parameters, a multidimensional parameter sequence related to the global typical failure mechanism of bearings is output, forming a spatial-temporal multiphysics feature database.

8. The method for evaluating the load-bearing capacity of an oil-impregnated bearing based on virtual simulation according to claim 1, characterized in that, Also includes: S9. Determine whether the extracted multidimensional parameter sequence meets the preset durability and failure criteria. If not, automatically optimize the modeling granularity and coupling ratio of the multiphysics sensitive area based on the feedback information to form an adaptive iteration of the model. S10 outputs high-fidelity multiphysics-dynamic load co-simulation analysis results after adaptive optimization, which are used for the prediction of durability and failure mechanism of oil-impregnated bearings under complex dynamic conditions and authoritative load-bearing capacity assessment.

Images