Bearing wear life prediction method and analysis system

Abstract

The invention relates to the technical field of bearing wear prediction, and particularly discloses a bearing wear life prediction method and analysis system.In the bearing wear simulation analysis process, wear loss, working conditions, simulation operation time and bearing geometric parameters are recorded under each time step length of simulation; further introducing a deep neural network model to carry out embedded coding and multivariable correlation learning on the simulation offline data of each time point so as to capture the bearing wear characteristics of each time point, and then carrying out dynamic propagation coding based on time dimension on the bearing wear characteristics of each time point. Therefore, the dynamic evolution rule of the bearing wear is excavated, and the intelligent prediction of the wear life of the bearing is realized. The method not only considers the real-time state of bearing wear, but also realizes continuous monitoring and dynamic analysis of the wear state of the bearing, so that the accuracy and reliability of prediction of the wear life of the bearing can be effectively improved.

Description

Technical Field

[0001] This application relates to the technical field of bearing wear prediction, and more specifically, to a bearing wear life prediction method and analysis system. Background Art

[0002] In the field of mechanical engineering, bearings, as key components of various rotating mechanical equipment, their performance and life are directly related to the reliability and efficiency of the entire system. With the development of industrial automation and intelligence, the monitoring of bearing wear status and life prediction become particularly important. Accurate wear life prediction not only helps prevent unexpected failures, reduce maintenance costs, but also optimize equipment maintenance plans and improve production efficiency.

[0003] Currently, the main methods for sliding bearing wear prediction mostly rely on model-based methods, such as finite element analysis (FEA), which evaluate the wear of bearings by directly performing numerical calculations and predictions. However, traditional finite element models have a long calculation time, making it difficult to meet the requirements of real-time prediction. Moreover, for different types of bearings, the model needs to be modified correspondingly, resulting in time-consuming, laborious, and high costs.

[0004] In response to this, the invention patent with the publication number of CN116721725A proposes a sliding bearing wear prediction and life prediction method, which establishes an accurate bearing material wear rate model based on experimental data, and uses a finite element model to generate bearing simulation wear data. Then, the bearing geometric parameters, working conditions, material parameters, and contact conditions in the bearing simulation wear data are used as inputs, and the wear amount of specific nodes is used as the output. The bearing simulation wear data is used to train a deep neural network model, so as to use the trained deep neural network model to predict the wear amount and life of the target sliding bearing, thus solving the problem of too long calculation time of traditional finite element models.

[0005] However, in the prior art, when using simulation offline data for bearing life prediction, it mainly focuses on the static characteristics under a set of conditions, and more relies on the data representation at discrete time points rather than continuous time manifolds, resulting in the failure to fully explore the dynamic characteristics of bearing wear over time, limiting the understanding of the long-term trend of the bearing movement wear process, and thus affecting the accuracy and reliability of the prediction.

[0006] Therefore, an optimized bearing wear life prediction method and analysis system are expected. Summary of the Invention

[0007] To solve the above technical problems, the present application is proposed. Embodiments of the present application provide a bearing wear life prediction method and an analysis system. During the bearing wear simulation analysis process, the wear amount, operating conditions, simulation running time, and bearing geometric parameters are recorded at each time step of the simulation. Further, a deep neural network model is introduced to perform embedded encoding and multi-variable correlation learning on the simulation offline data at each time point to capture the bearing wear characteristics at each time point. Furthermore, by performing dynamic propagation encoding based on the time dimension on the bearing wear characteristics at each time point, the dynamic evolution law of bearing wear is mined, thereby realizing the intelligent prediction of the bearing wear life. In this way, not only the immediate state of bearing wear is considered, but also the continuous monitoring and dynamic analysis of the bearing wear state are realized, thereby effectively improving the accuracy and reliability of bearing wear life prediction.

[0008] According to one aspect of the present application, a bearing wear life prediction method is provided, which includes:

[0009] Fitting a material wear rate model based on material wear data to establish a bearing wear finite element model;

[0010] Performing multi-level value-taking and combination on the input conditions to correct the parameters of the bearing wear finite element model, and performing numerical simulation based on the corrected bearing wear finite element model to obtain simulation wear data;

[0011] Generating offline data based on the input vector and output vector obtained by processing the simulation wear data;

[0012] Performing bearing wear life prediction on the target sliding bearing based on the offline data to obtain a prediction result;

[0013] In the above bearing wear life prediction method, performing bearing wear life prediction on the target sliding bearing based on the offline data to obtain a prediction result includes: extracting the time queues of the wear amount, operating conditions, simulation running time, and bearing geometric parameters from the offline data; performing multi-variable correlation encoding on the time queues of the wear amount, operating conditions, simulation running time, and bearing geometric parameters to obtain a time queue of multi-variable fully connected encoding vectors; performing feature dynamic propagation encoding based on gated writing on the time queue of the multi-variable fully connected encoding vectors to obtain a bearing wear simulation time series dynamic propagation encoding vector; and generating the prediction result based on the bearing wear simulation time series dynamic propagation encoding vector.

[0014] Preferably, a multivariate correlation encoding is performed on the time series of the wear amount, operating conditions, simulation running time, and bearing geometric parameters to obtain a time series of multivariate fully connected encoding vectors, including: performing embedding encoding on the wear amount, the operating conditions, the simulation running time, and the bearing geometric parameters respectively to obtain a time series of wear amount embedding encoding vectors, operating condition embedding encoding vectors, simulation running time embedding encoding vectors, and bearing geometric parameter embedding encoding vectors; inputting each wear amount embedding encoding vector, operating condition embedding encoding vector, simulation running time embedding encoding vector, and bearing geometric parameter embedding encoding vector in the time series of the wear amount embedding encoding vectors, operating condition embedding encoding vectors, simulation running time embedding encoding vectors, and bearing geometric parameter embedding encoding vectors into a multivariate time step encoder based on a fully connected layer to obtain the time series of the multivariate fully connected encoding vectors.

[0015] Preferably, a feature dynamic propagation encoding based on gated writing is performed on the time series of the multivariate fully connected encoding vectors to obtain a bearing wear simulation time series dynamic propagation encoding vector, including: performing a time series forward propagation encoding on the time series of the multivariate fully connected encoding vectors to obtain a bearing wear simulation multivariate time series initial propagation encoding vector; compensating and optimizing the bearing wear simulation multivariate time series initial propagation encoding vector based on the private features of each multivariate fully connected encoding vector in the time series of the multivariate fully connected encoding vectors with respect to the bearing wear simulation multivariate time series initial propagation encoding vector to obtain the bearing wear simulation time series dynamic propagation encoding vector.

[0016] Preferably, performing a time series forward propagation encoding on the time series of the multivariate fully connected encoding vectors to obtain a bearing wear simulation multivariate time series initial propagation encoding vector includes: inputting the time series of the multivariate fully connected encoding vectors into a sequence inference module based on a forward LSTM model to obtain the bearing wear simulation multivariate time series initial propagation encoding vector.

[0017] Preferably, based on the private features of each multivariate fully-connected encoded vector in the time queue of the multivariate fully-connected encoded vectors with respect to the initial propagation encoded vector of the bearing wear simulation multivariate time series, compensating and optimizing the initial propagation encoded vector of the bearing wear simulation multivariate time series to obtain the dynamic propagation encoded vector of the bearing wear simulation time series, including: calculating the private features of each multivariate fully-connected encoded vector in the time queue of the multivariate fully-connected encoded vectors with respect to the initial propagation encoded vector of the bearing wear simulation multivariate time series to obtain a time queue of the private feature encoded vectors of the bearing wear simulation multivariate time series nodes; performing gated writing on each private feature encoded vector of the bearing wear simulation multivariate time series nodes in the time queue of the private feature encoded vectors of the bearing wear simulation multivariate time series nodes to obtain a time queue of the gated-written private feature encoded vectors of the bearing wear simulation multivariate time series nodes; fusing the time queue of the gated-written private feature encoded vectors of the bearing wear simulation multivariate time series nodes and the initial propagation encoded vector of the bearing wear simulation multivariate time series to obtain the dynamic propagation encoded vector of the bearing wear simulation time series.

[0018] Preferably, performing gated writing on each private feature encoded vector of the bearing wear simulation multivariate time series nodes in the time queue of the private feature encoded vectors of the bearing wear simulation multivariate time series nodes to obtain a time queue of the gated-written private feature encoded vectors of the bearing wear simulation multivariate time series nodes, including: inputting the private feature encoded vector of the bearing wear simulation multivariate time series nodes into a probability unit based on the Softmax function to obtain a probability private feature encoded vector of the bearing wear simulation multivariate time series nodes; performing gated screening on the probability private feature encoded vector of the bearing wear simulation multivariate time series nodes based on a preset gated threshold to obtain a private feature mask weight vector of the bearing wear simulation multivariate time series nodes; calculating the element-wise product between the private feature mask weight vector of the bearing wear simulation multivariate time series nodes and the private feature encoded vector of the bearing wear simulation multivariate time series nodes to obtain the gated-written private feature encoded vector of the bearing wear simulation multivariate time series nodes.

[0019] Preferably, fusing the time queue of the gated-written private feature encoded vectors of the bearing wear simulation multivariate time series nodes and the initial propagation encoded vector of the bearing wear simulation multivariate time series to obtain the dynamic propagation encoded vector of the bearing wear simulation time series, including: concatenating the time queue of the gated-written private feature encoded vectors of the bearing wear simulation multivariate time series nodes and the initial propagation encoded vector of the bearing wear simulation multivariate time series to obtain the dynamic propagation encoded vector of the bearing wear simulation time series.

[0020] Preferably, generating the prediction result based on the bearing wear simulation time-series dynamic propagation coding vector includes: inputting the bearing wear simulation time-series dynamic propagation coding vector into a life prediction module based on a decoder to obtain a decoded value of the bearing wear life as the prediction result.

[0021] According to another aspect of the present application, there is provided a bearing wear life prediction and analysis system, which includes:

[0022] A data extraction module, configured to extract a time queue of wear amount, operating conditions, simulation running time, and bearing geometric parameters from the offline data of the bearing wear finite element model;

[0023] A multi-variable association coding module, configured to perform multi-variable association coding on the time queue of the wear amount, operating conditions, simulation running time, and bearing geometric parameters to obtain a time queue of multi-variable fully connected coding vectors;

[0024] A feature dynamic propagation coding module, configured to perform feature dynamic propagation coding based on gated writing on the time queue of the multi-variable fully connected coding vectors to obtain a bearing wear simulation time-series dynamic propagation coding vector;

[0025] A prediction result generation module, configured to generate the prediction result based on the bearing wear simulation time-series dynamic propagation coding vector.

[0026] The present application has at least the following technical effects:

[0027] Compared with the prior art, the bearing wear life prediction method and analysis system provided by the present application record the wear amount, operating conditions, simulation running time, and bearing geometric parameters at each time step during the bearing wear simulation analysis, and further introduce a deep neural network model to perform embedding coding and multi-variable association learning on the simulation offline data at each time point to capture the bearing wear characteristics at each time point. Furthermore, by performing dynamic propagation coding based on the time dimension on the bearing wear characteristics at each time point, the dynamic evolution law of bearing wear is excavated, thereby realizing the intelligent prediction of the bearing wear life. The present application not only considers the immediate state of bearing wear but also realizes continuous monitoring and dynamic analysis of the bearing wear state, thus effectively improving the accuracy and reliability of bearing wear life prediction. Description of the Drawings

[0028] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0029] Figure 1 It is a flowchart of a bearing wear life prediction method according to an embodiment of the present application.

[0030] Figure 2 It is a schematic diagram of data flow of a bearing wear life prediction method according to an embodiment of the present application.

[0031] Figure 3 It is a flowchart of sub-step S2 of a bearing wear life prediction method according to an embodiment of the present application.

[0032] Figure 4 It is a flowchart of sub-step S3 of a bearing wear life prediction method according to an embodiment of the present application.

[0033] Figure 5 It is a flowchart of sub-step S32 of a bearing wear life prediction method according to an embodiment of the present application.

[0034] Figure 6 It is a block diagram of a bearing wear life prediction analysis system according to an embodiment of the present application. Detailed Embodiments

[0035] As shown in the present application and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "including" and "comprising" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0036] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0037] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or below do not necessarily need to be executed precisely in sequence. On the contrary, according to the need, various steps can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several steps can be removed from these processes.

[0038] Hereinafter, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0039] It should be noted that in the present application, all actions of obtaining data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where the data is located and obtaining the authorization given by the owner of the corresponding device.

[0040] As mentioned in the above background art, Patent CN116721725A proposes a sliding bearing wear prediction and life prediction method, which includes: fitting a material wear rate model based on material wear data to establish a bearing wear finite element model; taking multi-level values and combinations of input conditions to correct the parameters of the bearing wear finite element model, and performing numerical simulation based on the corrected bearing wear finite element model to obtain simulation wear data; generating offline data based on the input vector and output vector obtained by processing the simulation wear data; predicting the bearing wear life of the target sliding bearing based on the offline data to obtain a prediction result.

[0041] The above solution solves the problem of too long calculation time of the traditional finite element model by establishing a bearing wear finite element model to perform bearing wear numerical simulation, so as to train a deep neural network model using the generated bearing simulation wear data, and then use the trained deep neural network model to predict the wear amount and life of the target sliding bearing.

[0042] However, in the prior art, when predicting the bearing life using simulation offline data, the static characteristics under a set of conditions are mainly focused on, and more reliance is placed on the data representation at discrete time points rather than the continuous time manifold. As a result, the dynamic characteristics of bearing wear evolving over time are not fully explored, limiting the understanding of the long-term trend of the bearing movement wear process, thus affecting the accuracy and reliability of the prediction. To address this technical problem, the present application proposes an optimized method for predicting the bearing wear life. During the process of bearing wear simulation analysis, the wear amount, operating conditions, simulation running time, and bearing geometric parameters are recorded at each time step of the simulation. Furthermore, a deep neural network model is introduced to perform embedding encoding and multi-variable correlation learning on the simulation offline data at each time point to capture the bearing wear characteristics at each time point. Then, through dynamic propagation encoding based on the time dimension for the bearing wear characteristics at each time point, the dynamic evolution law of bearing wear is mined, thereby realizing the intelligent prediction of the bearing wear life. In this way, not only the immediate state of bearing wear is considered, but also the continuous monitoring and dynamic analysis of the bearing wear state are achieved, thus effectively improving the accuracy and reliability of the bearing wear life prediction.

[0043] In the above method for predicting the sliding bearing wear and life, first, the material wear rate is modeled using the material wear data obtained from experiments or existing literature. This step involves selecting an appropriate mathematical model to describe the wear behavior of the material under specific operating conditions. Common wear models include Archard's equation, linear wear model, power-law wear model, etc. These models usually contain several core parameters, such as contact stress, relative sliding distance, and material hardness, etc., to quantify the relationship between the wear amount and the above factors. Once the wear rate model is determined, it can be integrated into the finite element analysis (FEA) software to simulate the interaction between the contact surfaces inside the sliding bearing and the accompanying material loss process. In this way, a preliminary finite element model of bearing wear is established.

[0044] To ensure the effectiveness and reliability of the model, a large amount of experimental data needs to be collected, and statistical methods are used to fit the model parameters. For example, the unknown parameters in the model can be adjusted by the least squares method or other optimization algorithms to make the model prediction values as close as possible to the actual measurement results. In addition, cross-validation techniques can be introduced to evaluate the generalization ability of the model to ensure that it is not only applicable to the data points within the training set but also can make reasonable predictions for external new samples. Once the wear rate model is determined, it can be integrated into the finite element analysis (FEA) software to simulate the interaction between the contact surfaces inside the sliding bearing and the accompanying material loss process. In this way, a preliminary finite element model of bearing wear is established.

[0045] To improve the accuracy and applicability of the model, it is necessary to consider various factors that may affect bearing wear and set different value levels for these factors. Operating load, rotational speed, temperature, lubricant type and quality, etc. are all important input variables. Each factor may affect the bearing wear rate individually or jointly. Therefore, it is particularly important to use the Design of Experiments (DOE) method to systematically select combinations of different levels for simulation experiments. This method can help identify which factors and the interaction effects between factors have the most significant impact on the wear results.

[0046] Specifically, a series of simulation tests can be planned through methods such as full factorial design, fractional factorial design, or response surface design. In each test, change the levels of one or more factors and record the corresponding output results, that is, the wear amounts of various parts of the bearing. Based on the comparison between these simulation outputs and the actually observed behavior, adjust the key parameters in the finite element model, such as material properties, friction coefficients, heat conduction characteristics, etc., to ensure that the model can more realistically reflect the bearing wear characteristics in the actual operating environment. After multiple iterations of optimization, a validated and parameter-corrected finite element model of bearing wear is finally obtained. This model can not only better capture the wear laws of bearings under complex working conditions but also provide a more accurate basis for subsequent numerical simulations.

[0047] Next, use the above-mentioned corrected and validated finite element model to perform a series of numerical simulations under different working conditions to generate detailed simulation wear data. Such simulations usually include multiple time steps, recording the stress distribution, deformation conditions, and corresponding wear amount changes of each part of the bearing at each moment. During the simulation process, not only static characteristics such as initial geometric dimensions and material properties should be concerned, but also dynamic characteristics such as the stress-strain response, temperature field distribution, and wear process evolving over time should be emphasized. In addition, by changing the boundary conditions (such as loading mode, lubrication state, etc.), the specific effects of different factors on the wear process can be further explored. All this information is recorded in detail to form a simulation database containing rich details. This database not only covers the steady-state behavior under a single working condition but also includes the transient response during the transition stage, providing valuable data support for in-depth understanding of the bearing working mechanism. By sorting and standardizing the data of a large number of simulation cases, the influence laws of different working conditions on the bearing wear life can be revealed, providing a scientific basis for formulating reasonable maintenance strategies. At the same time, these simulation data also provide high-quality basic materials for the subsequent training of machine learning models, helping to improve the prediction accuracy and reliability.

[0048] Next, extract useful information from the simulation wear data to construct input vectors and output vectors for subsequent analysis. The input vectors mainly include key parameters affecting wear, such as initial geometric dimensions, working load, rotational speed, temperature, lubricant type, etc.; while the output vectors reflect the wear amount and other related performance indicators at each time point. By sorting and standardizing the data of a large number of simulation cases, a structured offline data is formed, providing high-quality basic materials for the subsequent training of deep learning models. To ensure the quality and consistency of the data, it is recommended to adopt a unified data format and coding rules to convert the original simulation results into a table form that is easy to process. Each entry represents an independent simulation test, which contains all relevant input and output variables. In addition, it is necessary to consider how to handle problems such as missing values and outliers to ensure the integrity and accuracy of the data. For those discontinuous or intermittent variables, such as sudden increases in transient loads, special treatment is required to ensure that their impact on the results is correctly reflected. Finally, according to actual needs, an appropriate time resolution and sampling frequency can be selected to balance the relationship between the data volume and the computational efficiency.

[0049] Specifically, Figure 1 is a flowchart of the bearing wear life prediction method according to an embodiment of the present application. Figure 2 is a schematic diagram of data flow of the bearing wear life prediction method according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the bearing wear life prediction method includes the steps of: S1, extracting a time queue of wear amount, working conditions, simulation running time, and bearing geometric parameters from the offline data; S2, performing multivariate correlation coding on the time queue of wear amount, working conditions, simulation running time, and bearing geometric parameters to obtain a time queue of multivariate fully connected coding vectors; S3, performing feature dynamic propagation coding based on gated writing on the time queue of the multivariate fully connected coding vectors to obtain a bearing wear simulation time series dynamic propagation coding vector; S4, generating the prediction result based on the bearing wear simulation time series dynamic propagation coding vector.

[0050] In the above-mentioned bearing wear life prediction method, in step S1, a time series of wear amount, operating conditions, simulation running time, and bearing geometric parameters is extracted from the off-line data. It should be understood that the wear amount directly reflects the physical wear state of the bearing; the operating conditions (such as temperature, load, speed, etc.) affect the wear rate; the simulation running time provides the time background for wear occurrence; and the bearing geometric parameters determine its structural characteristics. Since bearing wear is a process that changes over time, as time goes by, various factors interact with each other and jointly affect the development of the bearing wear state. Therefore, in the technical solution of this application, by recording the wear amount, operating conditions, simulation running time, and bearing geometric parameters at each time point during the simulation process, the dynamic evolution law of bearing wear can be more comprehensively understood, thereby improving the prediction reliability of the remaining service life of the bearing.

[0051] Specifically, first, through finite element analysis (FEA) or other simulation means, the wear changes at different positions and time periods can be calculated. To track the wear progress of each part, specific detection points can be defined, and the wear values at these points are arranged in chronological order to form a time series. Such a time series can reveal the development trend of wear and provide important clues about the wear mechanism.

[0052] Next, since the operating conditions cover all external factors affecting the bearing operation, including temperature, pressure, speed, load type (static or dynamic), lubricant type and its quality, etc., each factor will have different degrees of influence on the bearing working performance. Therefore, it is very important to record the specific operating conditions at specific detection points during the simulation process in detail before data analysis. These information are collected and sorted in a standardized way. For example, a data table containing all possible variables is established, with each row representing a specific detection point and each column corresponding to different operating parameters. For non-continuous or intermittent variables, such as a suddenly increased transient load, special treatment is required to ensure that its impact on the results is correctly reflected.

[0053] The simulation running time refers to the time length from the simulation starting point to each specific detection point. A shorter simulation time may not be able to capture the long-term wear effect, while a longer simulation time may lead to excessive consumption of computing resources. Therefore, choosing an appropriate simulation running time is crucial for balancing prediction accuracy and computing efficiency. In actual operation, a segmented simulation strategy can be adopted, that is, first perform short-term intensive sampling and then gradually extend to a longer time span, which can not only ensure data quality but also improve efficiency.

[0054] The geometric structure of the bearing directly affects the stress conditions and wear patterns. Common geometric parameters include diameter, width, radius of curvature, fit clearance, etc. When it comes to different types or specifications of bearings, there may be significant differences among these parameters. Therefore, a complete set of geometric descriptions is defined separately for each specific case, and all geometric parameters are converted into a time series in a unified format. This includes not only the dimensional information in the initial state but also tracking the changes that occur over time and during the wear process, such as the shape changes of the journal or bearing housing due to wear, which in turn affect the overall mechanical properties.

[0055] Finally, all the above types of offline data are integrated to form a coherent time queue. The positional relationship of each item of data needs to be reasonably arranged according to the actual physical meaning. The wear amount, operating conditions, simulation running time, and geometric parameters at the same moment should be placed together to maintain logical relevance. In this way, the wear characteristics of the bearing throughout its entire life cycle can be more comprehensively grasped, laying a good foundation for further in-depth analysis and prediction.

[0056] In the above bearing wear life prediction method, in step S2, a multivariate correlation encoding is performed on the time queue of the wear amount, operating conditions, simulation running time, and bearing geometric parameters to obtain a time queue of multivariate fully connected encoding vectors. It should be understood that in this application, considering that there is a complex non-linear correlation relationship between the wear amount of the bearing and the operating conditions, simulation running time, and bearing geometric parameters under the same time state, therefore, in order to effectively capture the internal connections and interactions between these parameters, this application further performs a multivariate correlation analysis on the wear amount, operating conditions, simulation running time, and bearing geometric parameters at each time point to better simulate the actual situation of bearing wear. Among them, Figure 3 is a flowchart of sub-step S2 of the bearing wear life prediction method according to an embodiment of the present application. As Figure 3 shown, step S2 includes the steps of: S21, respectively performing embedding encoding on the wear amount, the operating conditions, the simulation running time, and the bearing geometric parameters to obtain a time queue of wear amount embedding encoding vectors, operating condition embedding encoding vectors, simulation running time embedding encoding vectors, and bearing geometric parameter embedding encoding vectors; S22, respectively inputting the wear amount embedding encoding vectors, operating condition embedding encoding vectors, simulation running time embedding encoding vectors, and bearing geometric parameter embedding encoding vectors in the time queue of the wear amount embedding encoding vectors, operating condition embedding encoding vectors, simulation running time embedding encoding vectors, and bearing geometric parameter embedding encoding vectors into a multivariate time step encoder based on a fully connected layer to obtain the time queue of the multivariate fully connected encoding vectors.

[0057] Specifically, in step S21, the wear amount, the working condition, the simulation running time, and the bearing geometric parameters are respectively embedded and encoded to obtain a time queue of wear amount embedded encoding vectors, working condition embedded encoding vectors, simulation running time embedded encoding vectors, and bearing geometric parameter embedded encoding vectors. It should be understood that considering that each parameter in the original data has different dimensions and numerical ranges, effective feature interaction cannot be directly performed. Therefore, in order to achieve effective fusion of information, the present application further processes the wear amount, the working condition, the simulation running time, and the bearing geometric parameters respectively through an embedding encoding technique to map multi-dimensional parameter information into a low-dimensional dense vector space, convert it into a unified numerical vector representation form, and form a time queue of wear amount embedded encoding vectors, working condition embedded encoding vectors, simulation running time embedded encoding vectors, and bearing geometric parameter embedded encoding vectors, thereby providing a basis for subsequent multi-variable correlation analysis and bearing wear state time series analysis.

[0058] Specifically, in step S22, each wear amount embedded encoding vector, working condition embedded encoding vector, simulation running time embedded encoding vector, and bearing geometric parameter embedded encoding vector in the time queue of wear amount embedded encoding vectors, working condition embedded encoding vectors, simulation running time embedded encoding vectors, and bearing geometric parameter embedded encoding vectors are respectively input into a multi-variable time step encoder based on a fully connected layer to obtain a time queue of multi-variable fully connected encoding vectors. Specifically, the present application uses a multi-variable time step encoder based on a fully connected layer to perform correlation learning on the wear amount embedded encoding vectors, working condition embedded encoding vectors, simulation running time embedded encoding vectors, and bearing geometric parameter embedded encoding vectors at each time step to capture multi-dimensional information of the bearing state at different time steps and generate a time queue of multi-variable fully connected encoding vectors. It should be understood that the fully connected layer is a deep learning network structure that allows each input node to be connected to all nodes in the next layer, thereby learning the potential interaction and dependence relationships between multi-dimensional features and enhancing the comprehensive understanding of the bearing wear state. In this way, the multi-dimensional parameter information at each time point can be effectively fused, so as to ensure that all relevant factors can be considered simultaneously in the subsequent bearing wear life prediction process.

[0059] In the above bearing wear life prediction method, in step S3, feature dynamic propagation encoding based on gated writing is performed on the time queue of the multivariate fully-connected encoding vectors to obtain a bearing wear simulation time-series dynamic propagation encoding vector. It should be understood that considering that the wear state of the bearing gradually accumulates over time, and this time-series cumulative effect of the wear state has an important impact on the remaining service life of the bearing. Therefore, in order to improve the accuracy of bearing wear life prediction, the present application further performs time-series dynamic propagation encoding on the time queue of the multivariate fully-connected encoding vectors to simulate the dynamic change of the bearing wear state over time, so as to more accurately predict the wear trend and remaining life of the bearing. Among them, Figure 4 is a flowchart of sub-step S3 of the bearing wear life prediction method according to an embodiment of the present application. As Figure 4 shown, step S3 includes steps: S31, performing time-series forward propagation encoding on the time queue of the multivariate fully-connected encoding vectors to obtain a bearing wear simulation multivariate time-series initial propagation encoding vector; S32, based on the private features of each multivariate fully-connected encoding vector in the time queue of the multivariate fully-connected encoding vectors relative to the bearing wear simulation multivariate time-series initial propagation encoding vector, compensating and optimizing the bearing wear simulation multivariate time-series initial propagation encoding vector to obtain the bearing wear simulation time-series dynamic propagation encoding vector.

[0060] Specifically, in a specific example of the present application, step S31 includes: inputting the time queue of the multivariate fully-connected encoding vectors into a sequence inference module based on a forward LSTM model to obtain the bearing wear simulation multivariate time-series initial propagation encoding vector, which is expressed by the formula:

[0061] X = {x1, x2,..., x i ,..., x n}

[0062] v h = LSTM({x1, x2,..., x i ,..., x n )

[0063] where X represents the time queue of the multivariate fully-connected encoding vectors, x1, x2, x i and x n respectively represent the 1st, 2nd, ith, and nth multivariate fully-connected encoding vectors in the set of the multivariate fully-connected encoding vectors, n is the number of feature vectors in the set of the multivariate fully-connected encoding vectors, LSTM(·) represents the forward LSTM model, and v h represents the bearing wear simulation multivariate time-series initial propagation encoding vector.

[0064] That is, considering that the LSTM model has significant advantages in time series data processing tasks and can effectively capture long-term dependencies in the data, therefore, in this application, a forward LSTM model is first adopted to perform forward propagation encoding on the time queue of the multivariate fully connected encoded vectors, so as to utilize the long-term memory ability of the LSTM model to capture the long-range dependencies between the bearing wear multivariate features at each time step, and generate a hidden state representation that synthesizes the bearing wear multivariate feature information of all time steps, that is, the initial propagation encoded vector of the bearing wear simulation multivariate time series.

[0065] Figure 5 It is a flowchart of sub-step S32 of the bearing wear life prediction method according to an embodiment of the present application. As Figure 5 shown, the step S32 includes steps: S321, calculating the private features of each multivariate fully connected encoded vector in the time queue of the multivariate fully connected encoded vectors relative to the initial propagation encoded vector of the bearing wear simulation multivariate time series to obtain a time queue of the bearing wear simulation multivariate time series node private feature encoded vectors; S322, performing gated writing on each bearing wear simulation multivariate time series node private feature encoded vector in the time queue of the bearing wear simulation multivariate time series node private feature encoded vectors to obtain a time queue of the gated written bearing wear simulation multivariate time series node private feature encoded vectors; S323, fusing the time queue of the gated written bearing wear simulation multivariate time series node private feature encoded vectors and the initial propagation encoded vector of the bearing wear simulation multivariate time series to obtain the dynamic propagation encoded vector of the bearing wear simulation time series.

[0066] More specifically, the step S321 is expressed by the formula:

[0067]

[0068] where g(·, ·) represents a private feature extraction function, represents position difference,

[0069] |·| represents taking the absolute value, conv 1×1 (·) represents point convolution, Sigmoid(·) represents the sigmoid activation function, W d is a weight parameter matrix, p i represents the bearing wear simulation multivariate time series node difference feature vector corresponding to the x i , and xb i represents the bearing wear simulation multivariate time series node private feature encoded vector corresponding to the x i .

[0070] That is, considering that the LSTM model mainly controls the flow of information through the gating mechanism, during the feature propagation process, the multivariate features of bearing wear at each time step may be lost or wrongly filtered. To address this, the present application further generates a time queue of the bearing wear simulation multivariate time series node private feature encoding vectors by calculating the private features of the original multivariate fully connected encoding vectors with respect to the bearing wear simulation multivariate time series initial propagation encoding vectors, so as to reflect the information loss during the feature propagation process.

[0071] More specifically, in a specific example of the present application, the step S322 includes: First, input the bearing wear simulation multivariate time series node private feature encoding vector into a probability unit based on the Softmax function to obtain a probability-bearing wear simulation multivariate time series node private feature encoding vector, which is expressed by the formula:

[0072]

[0073] where a i represents the probability-bearing wear simulation multivariate time series node private feature encoding vector corresponding to the xb i , represents matrix multiplication operation, W i and b i respectively represent the weight matrix and bias term of the probability unit, softmax(·) is the normalized exponential function,

[0074] is the inverse temperature coefficient of the Softmax function.

[0075] That is, considering that the private features of each bearing wear simulation multivariate time series node may contain important feature information and may also be subject to noise interference, therefore, in order to further implement the screening and filtering of the private features, the present application first evaluates the importance of each part of the features in the bearing wear simulation multivariate time series node private feature encoding vector through a probability unit based on the Softmax function, and converts it into a probability distribution, which can be intuitively understood as the importance score of different parts of the private features.

[0076] Then, based on a preset gating threshold, perform gating screening on the probability-bearing wear simulation multivariate time series node private feature encoding vector to obtain a bearing wear simulation multivariate time series node private feature mask weight vector, which is expressed by the formula:

[0077]

[0078] where mask(·) is the gating mask function, τ is the preset gating threshold, and w i is the xbi The corresponding bearing wear simulation multivariate time series node private feature mask weight vector.

[0079] That is, by performing gated masking on the obtained probability distribution, the corresponding mask weights are generated to guide the selection and filtering of private features to determine whether they need to be written into the bearing wear simulation multivariate time series initial propagation coding vector.

[0080] Finally, calculate the element-wise multiplication between the bearing wear simulation multivariate time series node private feature mask weight vector and the bearing wear simulation multivariate time series node private feature coding vector to obtain the gated write bearing wear simulation multivariate time series node private feature coding vector, which is expressed by the formula:

[0081] Y = {xb1·w1, xb2·w2, …, xb i ·w i , …, xb n ·w n}

[0082] Where Y is the time queue of the gated write bearing wear simulation multivariate time series node private feature coding vector, xb1, xb2, and xb n respectively represent the bearing wear simulation multivariate time series node private feature coding vectors corresponding to the x1, x2, and x n , and w1, w2, and w n respectively are the bearing wear simulation multivariate time series node private feature mask weight vectors corresponding to the xb1, xb2, and xb n .

[0083] That is, the generated mask weights are used to weight the corresponding bearing wear simulation multivariate time series node private feature coding vectors, thereby realizing the gated write of the bearing wear multivariate private features at each time step to ensure that only the most representative and relevant private information is retained and transmitted.

[0084] More specifically, in a specific example of the present application, the step S323 includes: concatenating the time queue of the gated write bearing wear simulation multivariate time series node private feature coding vector and the bearing wear simulation multivariate time series initial propagation coding vector to obtain the bearing wear simulation time series dynamic propagation coding vector, which is expressed by the formula:

[0085] V = Concat{v h ; Y}

[0086] Where Concat{·; ·} represents the concatenation operation, and V represents the bearing wear simulation time series dynamic propagation coding vector.

[0087] That is, the time queue of the gated write bearing wear simulation multivariate time series node private feature encoding vectors and the bearing wear simulation multivariate time series initial propagation encoding vectors are spliced and fused to simultaneously retain the overall change trend of the bearing wear state and the detailed information of each time step, obtaining the bearing wear simulation time series dynamic propagation encoding vector, thereby ensuring the integrity and accuracy of information during the feature propagation process.

[0088] In the above bearing wear life prediction method, in step S4, based on the bearing wear simulation time series dynamic propagation encoding vector, the prediction result is generated. In a specific example of the present application, step S4 includes: inputting the bearing wear simulation time series dynamic propagation encoding vector into a life prediction module based on a decoder to obtain a decoded value of the bearing wear life as the prediction result. It should be understood that the decoder can learn the complex patterns and temporal dependencies in the bearing wear simulation time series dynamic propagation encoding vector through a multi-layer neural network (such as RNN, LSTM, or GRU), and based on the temporal cumulative effect of this bearing wear state, generate a corresponding bearing wear life prediction value, thereby achieving an accurate prediction of the bearing wear life. During the training process, the decoder optimizes its internal parameters by minimizing the difference between the predicted value and the true value, thereby ensuring its high precision and good generalization ability, providing real-time and accurate wear predictions for bearing operation and maintenance personnel to support preventive maintenance decisions.

[0089] This application considers that each multivariate fully connected encoding vector in the time queue of the multivariate fully connected encoding vectors respectively represents the embedded encoding fully connected interaction encoding features of {wear amount, operating conditions, simulation running time, bearing geometric parameters} at each predetermined time point. When performing feature dynamic propagation based on gated write, the feature heterogeneity and gated write sparsity at different time nodes will cause the aggregation feature instance neighborhood of the bearing wear simulation time series dynamic propagation encoding vector to be sparsified, thereby affecting the accuracy of the decoded value of the bearing wear life obtained by inputting it into the life prediction module based on the decoder.

[0090] Based on this, before inputting the bearing wear simulation time series dynamic propagation encoding vector into the life prediction module based on the decoder, the bearing wear simulation time series dynamic propagation encoding vector is first optimized, including the steps of:

[0091] Performing linear difference measurement and non-linear difference measurement on the element features at any two positions in the bearing wear simulation time series dynamic propagation encoding vector to obtain a first bearing wear simulation time series dynamic difference measurement characterization matrix and a second bearing wear simulation time series dynamic difference measurement characterization matrix, expressed as:

[0092]

[0093]

[0094] v i ,v j ∈V

[0095] where v i and v j respectively represent the element features at any two positions in the bearing wear simulation time-series dynamic propagation coding vector, w1 represents the first weighted hyperparameter, w2 represents the second weighted hyperparameter, V represents the bearing wear simulation time-series dynamic propagation coding vector, represents the value at the (i, j) position in the first bearing wear simulation time-series dynamic difference metric characterization matrix, represents the value at the (i, j) position in the second bearing wear simulation time-series dynamic difference metric characterization matrix;

[0096] Based on the first bearing wear simulation time-series dynamic difference metric characterization matrix and the second bearing wear simulation time-series dynamic difference metric characterization matrix, construct a bearing wear simulation time-series dynamic multi-dimensional difference metric coding matrix, denoted as:

[0097] D = D1 ⊙ D2

[0098] where ⊙ represents element-wise multiplication, and D represents the bearing wear simulation time-series dynamic multi-dimensional difference metric coding matrix;

[0099] Perform feature ontology self-association on the feature distribution of the bearing wear simulation time-series dynamic propagation coding vector to obtain a bearing wear simulation time-series dynamic ontology self-association matrix, denoted as:

[0100]

[0101] where T represents the transpose of the vector, and M represents the bearing wear simulation time-series dynamic ontology self-association matrix;

[0102] Using the bearing wear simulation time-series dynamic multi-dimensional difference metric coding matrix as the modulation target matrix, perform mapping modulation on the bearing wear simulation time-series dynamic propagation coding vector to obtain a bearing wear simulation time-series dynamic base projection conversion coding vector, denoted as:

[0103]

[0104] where V1 represents the bearing wear simulation time-series dynamic base projection conversion coding vector;

[0105] Construct a hierarchical topological target modulation matrix by combining the bearing wear simulation time-series dynamic multi-dimensional difference metric coding matrix and the bearing wear simulation time-series dynamic multi-dimensional difference metric coding matrix, and perform mapping modulation on the bearing wear simulation time-series dynamic base projection conversion coding vector based on the hierarchical topological target modulation matrix to obtain a bearing wear simulation time-series dynamic target coding vector, expressed as:

[0106]

[0107] where, M s represents the hierarchical topological target modulation matrix, and V2 represents the bearing wear simulation time-series dynamic target coding vector;

[0108] Fuse the bearing wear simulation time-series dynamic target coding vector and the bearing wear simulation time-series dynamic propagation coding vector to obtain an optimized bearing wear simulation time-series dynamic propagation coding vector, expressed as:

[0109] V′ = V2 ☉ V

[0110] where, V′ represents the optimized bearing wear simulation time-series dynamic propagation coding vector.

[0111] Correspondingly, in this preferred embodiment, by constructing an oriented projection conversion representation framework of difference metric representation, a hierarchical topological structure re-projection conversion is implemented for the generation process of all isomorphic instances associated with the body of the bearing wear simulation time-series dynamic propagation coding vector. Furthermore, a linkage fusion kernel offset compensation mechanism is adopted to offset the interference effect of the linkage mismatch negative interference factor, thereby enhancing the criterion instantiation intensity of the bearing wear simulation time-series dynamic propagation coding vector under similarity constraints, so as to improve the fine-grained mapping accuracy of the bearing wear simulation time-series dynamic propagation coding vector in the decoding process, and thus improve the accuracy of the decoding value of the bearing wear life obtained by inputting the bearing wear simulation time-series dynamic propagation coding vector into the life prediction module based on the decoder.

[0112] After obtaining the prediction result, it is also necessary to go through detailed verification and calibration. By comparing with the actual measurement data, the accuracy of the prediction model can be evaluated, and any possible deviations or anomalies can be identified. If a large difference is found between the predicted value and the actual situation, it is necessary to return to the modeling stage, re-examine and adjust the model parameters or structure until the prediction accuracy reaches the expected level. This iterative optimization process ensures the reliability and practicality of the final output.

[0113] The verified prediction results can provide important information about the future wear trend of the bearing. Using this information, a more scientific and reasonable maintenance plan can be formulated. For example, based on the predicted wear rate and remaining service life, the optimal maintenance time window can be determined to avoid premature or late maintenance. For critical equipment, advance planning can effectively reduce unplanned downtime and lower operating costs. At the same time, the working conditions can also be adjusted based on the prediction results, such as appropriately reducing the load, optimizing the lubrication scheme, etc., to delay the wear process and extend the actual service life of the bearing.

[0114] In summary, the bearing wear life prediction method based on the embodiments of the present application is elucidated. During the bearing wear simulation analysis process, the wear amount, working conditions, simulation running time, and bearing geometric parameters are recorded at each time step of the simulation. Further, a deep neural network model is introduced to perform embedded coding and multi-variable correlation learning on the simulation offline data at each time point to capture the bearing wear characteristics at each time point. Furthermore, through dynamic propagation coding based on the time dimension for the bearing wear characteristics at each time point, the dynamic evolution law of bearing wear is mined, thereby realizing the intelligent prediction of the bearing wear life. In this way, not only the immediate state of bearing wear is considered, but also the continuous monitoring and dynamic analysis of the bearing wear state are realized, thus effectively improving the accuracy and reliability of bearing wear life prediction.

[0115] Furthermore, a bearing wear life prediction and analysis system is also provided.

[0116] Figure 6 For the block diagram of the bearing wear life prediction and analysis system according to the embodiments of the present application. As Figure 6 shown, the bearing wear life prediction and analysis system 100 according to the embodiments of the present application includes: a data extraction module 110, configured to extract a time queue of the wear amount, working conditions, simulation running time, and bearing geometric parameters from the offline data of the bearing wear finite element model; a multi-variable correlation coding module 120, configured to perform multi-variable correlation coding on the time queue of the wear amount, working conditions, simulation running time, and bearing geometric parameters to obtain a time queue of multi-variable fully connected coding vectors; a feature dynamic propagation coding module 130, configured to perform feature dynamic propagation coding based on gated writing on the time queue of the multi-variable fully connected coding vectors to obtain a bearing wear simulation time series dynamic propagation coding vector; and a prediction result generation module 140, configured to generate the prediction result based on the bearing wear simulation time series dynamic propagation coding vector.

[0117] The specific operations of each module in the above bearing wear life prediction and analysis system have been described in detail in the description of the bearing wear life prediction method above with reference to Figures 1 to 5 and thus, the repeated description thereof will be omitted.

[0118] The basic principles of the present invention have been described above in connection with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present invention are only examples and not limitations. It cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitating understanding, rather than limitations. The above details do not limit the present invention to necessarily adopt the above specific details for implementation.

[0119] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0120] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced by the present invention. Any associated drawing reference signs in the claims should not be regarded as limiting the claimed rights.

[0121] In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0122] Finally, it should be noted that the above description has been given for purposes of illustration and description. In addition, the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting the wear life of a bearing, comprising: Fitting a material wear rate model based on material wear data to establish a finite element model of bearing wear; Taking multi-level values and combinations of input conditions to correct the parameters of the finite element model of bearing wear, and performing numerical simulation based on the corrected finite element model of bearing wear to obtain simulation wear data; generating offline data based on the input vector and output vector obtained by processing the simulation wear data; predicting the bearing wear life of the target sliding bearing based on the offline data to obtain a prediction result. It is characterized in that predicting the bearing wear life of the target sliding bearing based on the offline data to obtain a prediction result includes: Extracting the time queues of wear amount, working conditions, simulation running time, and bearing geometric parameters from the offline data; Performing multi-variable correlation encoding on the time queues of wear amount, working conditions, simulation running time, and bearing geometric parameters to obtain a time queue of multi-variable fully connected encoding vectors; Performing feature dynamic propagation encoding based on gated writing on the time queue of the multi-variable fully connected encoding vectors to obtain a bearing wear simulation time-series dynamic propagation encoding vector; Generating the prediction result based on the bearing wear simulation time-series dynamic propagation encoding vector.

2. The bearing wear life prediction method according to claim 1, wherein Performing multi-variable correlation encoding on the time queues of wear amount, working conditions, simulation running time, and bearing geometric parameters to obtain a time queue of multi-variable fully connected encoding vectors, including: Performing embedding encoding on the wear amount, the working conditions, the simulation running time, and the bearing geometric parameters respectively to obtain a time queue of wear amount embedding encoding vectors, working condition embedding encoding vectors, simulation running time embedding encoding vectors, and bearing geometric parameter embedding encoding vectors; Inputting each wear amount embedding encoding vector, working condition embedding encoding vector, simulation running time embedding encoding vector, and bearing geometric parameter embedding encoding vector in the time queues of wear amount embedding encoding vectors, working condition embedding encoding vectors, simulation running time embedding encoding vectors, and bearing geometric parameter embedding encoding vectors into a multi-variable time step encoder based on a fully connected layer to obtain the time queue of the multi-variable fully connected encoding vectors.

3. The bearing wear life prediction method according to claim 2, wherein Performing feature dynamic propagation encoding based on gated writing on the time queue of the multi-variable fully connected encoding vectors to obtain a bearing wear simulation time-series dynamic propagation encoding vector, including: Performing time-series forward propagation encoding on the time queue of the multi-variable fully connected encoding vectors to obtain a bearing wear simulation multi-variable time-series initial propagation encoding vector; Compensating and optimizing the bearing wear simulation multi-variable time-series initial propagation encoding vector based on the private features of each multi-variable fully connected encoding vector in the time queue of the multi-variable fully connected encoding vectors relative to the bearing wear simulation multi-variable time-series initial propagation encoding vector to obtain the bearing wear simulation time-series dynamic propagation encoding vector.

4. The bearing wear life prediction method according to claim 3, wherein Performing time-series forward propagation encoding on the time queue of the multi-variable fully connected encoding vectors to obtain a bearing wear simulation multi-variable time-series initial propagation encoding vector, including: Input the time queue of the multi-variable fully-connected encoded vectors into the sequence inference module based on the forward LSTM model to obtain the initial propagation encoded vectors of the multi-variable time series for bearing wear simulation.

5. The bearing wear life prediction method according to claim 4, wherein Based on the private features of each multi-variable fully-connected encoded vector in the time queue of the multi-variable fully-connected encoded vectors with respect to the initial propagation encoded vectors of the multi-variable time series for bearing wear simulation, compensate and optimize the initial propagation encoded vectors of the multi-variable time series for bearing wear simulation to obtain the dynamic propagation encoded vectors of the time series for bearing wear simulation, including: Calculate the private features of each multi-variable fully-connected encoded vector in the time queue of the multi-variable fully-connected encoded vectors with respect to the initial propagation encoded vectors of the multi-variable time series for bearing wear simulation to obtain the time queue of the private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation; Perform gated writing on each private feature encoded vector of the nodes of the multi-variable time series for bearing wear simulation in the time queue of the private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation to obtain the time queue of the gated-written private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation; Fuse the time queue of the gated-written private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation and the initial propagation encoded vectors of the multi-variable time series for bearing wear simulation to obtain the dynamic propagation encoded vectors of the time series for bearing wear simulation.

6. The bearing wear life prediction method according to claim 5, characterized in that Perform gated writing on each private feature encoded vector of the nodes of the multi-variable time series for bearing wear simulation in the time queue of the private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation to obtain the time queue of the gated-written private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation, including: Input the private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation into the probability unit based on the Softmax function to obtain the probability-private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation; Based on the preset gating threshold, perform gating screening on the probability-private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation to obtain the private feature mask weight vectors of the nodes of the multi-variable time series for bearing wear simulation; Calculate the element-wise multiplication between the private feature mask weight vectors of the nodes of the multi-variable time series for bearing wear simulation and the private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation to obtain the gated-written private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation.

7. The bearing wear life prediction method according to claim 6, wherein Fuse the time queue of the gated-written private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation and the initial propagation encoded vectors of the multi-variable time series for bearing wear simulation to obtain the dynamic propagation encoded vectors of the time series for bearing wear simulation, including: Cascade the time queue of the gated-written private feature encoded vectors of the nodes of the multi-variable time series for bearing wear simulation and the initial propagation encoded vectors of the multi-variable time series for bearing wear simulation to obtain the dynamic propagation encoded vectors of the time series for bearing wear simulation.

8. The bearing wear life prediction method according to claim 7, wherein Generate the prediction result based on the dynamic propagation encoded vectors of the time series for bearing wear simulation, including: Input the bearing wear simulation time-series dynamic propagation coding vector into the decoder-based life prediction module to obtain the decoded value of the bearing wear life as the prediction result.

9. A bearing wear life prediction and analysis system, which can implement the bearing wear life prediction method described in any one of claims 1-8, characterized in that, It includes: A data extraction module for extracting the time series of wear amount, working conditions, simulation running time, and bearing geometric parameters from the offline data of the bearing wear finite element model; A multi-variable correlation coding module for performing multi-variable correlation coding on the time series of the wear amount, working conditions, simulation running time, and bearing geometric parameters to obtain the time series of multi-variable fully connected coding vectors; A feature dynamic propagation coding module for performing feature dynamic propagation coding based on gated writing on the time series of the multi-variable fully connected coding vectors to obtain the bearing wear simulation time-series dynamic propagation coding vector; A prediction result generation module for generating a prediction result based on the bearing wear simulation time-series dynamic propagation coding vector.

Images