A dual-variable wet clutch online remaining life prediction method and system

Abstract

This invention discloses a method and system for predicting the remaining service life (RUL) of a dual-variable wet clutch online. The method includes: acquiring real-time friction torque data of the clutch and preprocessing the data to extract key performance characteristic parameters; constructing a dual-variable inverse Gaussian degradation model based on these parameters; and dynamically predicting the RUL of the clutch online using this model. Specifically, the parameters of the dual-variable inverse Gaussian degradation model are estimated using the Markov chain Monte Carlo (MCMC) method, and the online prediction of the RUL is achieved through extensive sampling using Monte Carlo simulation. This invention enables flexible fitting of the degradation process and dynamic, accurate online prediction of the remaining service life, providing precise model support for condition-based decision-making, optimizing system health management and maintenance plans, and reducing unexpected failures.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of combining equipment reliability engineering and stochastic processes, and in particular to a method and system for predicting the online remaining life of a dual-variable wet clutch. Background Technology

[0002] Wet clutches are widely used power transmission devices, commonly found in automatic transmissions, heavy machinery, and high-performance vehicles. With technological advancements and increasing demands, the performance and reliability of wet clutches have become increasingly important. Wet clutches transmit and switch power by controlling the engagement and disengagement of friction plates and steel plates, often operating under high loads and for extended periods. Therefore, under high loads and frequent use, the accumulated friction and heat effects lead to gradual wear of the clutch's friction elements, reducing its torque transmission capacity and severely impacting its normal operating quality, reliability, and the lifespan and safety of the vehicle's transmission system. Therefore, effectively monitoring the degradation data of wet clutches, i.e., performance characteristics (PC), can not only effectively reveal the degradation patterns of wet clutches but also provide a reliable historical information basis for predicting effective remaining useful life (RUL). By establishing accurate predictive models, the lifespan and reliability of wet clutches can be extended, reducing maintenance costs during the service life phase.

[0003] Degradation analysis, including degradation modeling and parameter estimation, has proven to be an important toolkit, especially for devices with limited testing time and sample sizes. A suitable degradation model is crucial for representing the degradation characteristics and reliability of clutches. Existing reliability analysis methods typically use univariate models to describe the degradation process of clutch systems. However, in practice, clutch degradation exhibits multivariate characteristics, and the variables may be correlated. Furthermore, most existing bivariate or multivariate models employ deterministic models or models based on simple stochastic processes, which have limitations in capturing the uncertainty and dynamic behavior of degradation. In recent years, inverse Gaussian (IG) process models have gained widespread attention due to their ability to effectively describe the stochasticity of system degradation processes. However, existing methods cannot adequately handle the changes in real-time dynamic data when dealing with multivariate degradation processes and real-time RUL prediction of complex systems, resulting in inaccurate predictions. Particularly for online RUL prediction, traditional estimation methods have low accuracy, making it impossible to provide accurate RUL for effective preventative maintenance arrangements, increasing the risk of sudden failures and maintenance costs. Existing methods lag behind in reliability assessment and preventative maintenance decision support, making it difficult to effectively schedule preventative maintenance based on accurate lifespan predictions. This leads to reduced system reliability and increased maintenance costs. A model is needed that can provide precise support for decision-making, optimize system health management and maintenance plans, and reduce unexpected failures. Summary of the Invention

[0004] To address the aforementioned technical issues, this invention proposes a bivariate wet clutch online remaining life prediction method and system. Based on real-time online characteristic parameter data of the clutch, a bivariate inverse Gaussian degradation model is constructed. Simultaneously considering the correlation and uncertainty of degradation variables, this method achieves flexible fitting of the degradation process and dynamic, accurate online prediction of remaining life, providing precise model support for situation-based decision-making, optimizing system health management and maintenance plans, and reducing unexpected failures.

[0005] On the one hand, to achieve the above objectives, the present invention provides a method for predicting the remaining life of a dual-variable wet clutch online, comprising:

[0006] Acquire real-time friction torque data of the clutch, and preprocess the real-time friction torque data to extract key performance characteristic parameters;

[0007] A bivariate inverse Gaussian degradation model is constructed based on the key performance characteristic parameters. The clutch RUL is then dynamically predicted online using this model. Specifically, the parameters of the bivariate inverse Gaussian degradation model are estimated using the Markov chain Monte Carlo (MCMC) method, and the clutch RUL is predicted online based on extensive sampling using the Monte Carlo simulation method.

[0008] Preferably, the real-time friction torque data of the clutch is acquired, and the real-time friction torque data is preprocessed, including:

[0009] Friction torque data is collected in real time by a speed and torque sensor, and the friction torque data is denoised and smoothed to eliminate noise and outliers.

[0010] Preferably, the method for extracting the key performance characteristic parameters is as follows:

[0011]

[0012] t e =t b -t a ;

[0013] In the formula, t e s is the engagement time; s is the number of friction pairs; p is the applied pressure; R o and R i These are the outer and inner radii of the friction component, respectively; t a and t d T(t) represents the start and end times of the engagement process, respectively. T(t) is the real-time friction torque, and μ is the friction coefficient COF.

[0014] Preferably, constructing the bivariate inverse Gaussian degradation model includes:

[0015] Performance characteristic parameters are obtained by using friction torque data, and it is assumed that random variables are controlled by probability distributions to obtain the probability density function of the degradation process. The performance characteristic parameters include the degradation process of the friction coefficient COF and engagement time.

[0016] By setting up n clutch units to conduct degradation tests under the same conditions, each unit i obtains m degradation observations. Given the latent variables of the degradation units, the posterior distribution of the latent variables is obtained, and finally the complete log-likelihood function of the bivariate inverse Gaussian degradation model is obtained.

[0017] Preferably, the probability density function of the degradation process is:

[0018] Let the extracted friction coefficient COF and bonding time be represented as Y1(t) and Y2(t), respectively, where t≥0, and they both follow an inverse Gaussian distribution, i.e., Y p (t)~IG(μΛ p (t), η p Λ p (t) 2 ); where p = 1, 2, representing two degradation processes respectively; μ is a common latent variable used to describe the dependency between the two degradation processes; Λ(t) is a time-dependent shape function, in power-law form. t and q are time and time powers, respectively; η describes the volatility of degradation, thus yielding the probability density function of the degradation process:

[0019]

[0020] In the formula, f(Y) p ) is the probability density function of the corresponding performance characteristic parameter.

[0021] Preferably, obtaining the probability density function of the degradation process includes:

[0022] Assume latent variable μ -1 TN(ω,κ) follows a truncated normal distribution -2 And statistically independent of the parameter η; then assume a given μ -1 If the degradation processes of the two performance parameters Y1(t) and Y2(t) are independent of each other, then the degradation process can be represented as Y p (t)|μ -1 ~IG(μΛ) p (t), η p Λ p (t) 2 Since p = 1 and 2, the probability density function of the degradation process is:

[0023] f(Y k (t)|μ -1 )=f(Y p );

[0024] The probability density function of the latent variable is:

[0025]

[0026] In the formula, π(μ) -1 Let φ be the probability density function of the latent variable, ω be the mean parameter, κ be the standard deviation parameter, φ be the probability density function of the standard normal distribution, and Φ be the cumulative distribution function of the standard normal distribution.

[0027] Preferably, the complete log-likelihood function of the bivariate inverse Gaussian degeneracy model is:

[0028]

[0029] In the formula, θ=(ω,κ,η1,η2,q1,q2) is the parameter set of the parameters to be estimated in the model, and P(·) is the latent variable of the given degenerate unit i. The likelihood function of the model, For m observations of the friction coefficient COF of unit i, For m observations of the engagement time of unit i, Let N be the common latent variable of the i-th clutch unit, and N be the number of clutch units.

[0030] Preferably, the dynamic online prediction of clutch RUL using the bivariate inverse Gaussian degradation model includes:

[0031] The Monte Carlo simulation method is used to perform iterative prediction of the RUL (Restricted Usage Limit) of the clutch system using time windows of the degradation data of the obtained performance parameters. Specifically:

[0032] S1. Input the known historical data of the current time t into the bivariate inverse Gaussian degradation model, and obtain the posterior distribution range of each parameter, including the mean and variance, based on the MCMC method;

[0033] S2. When the latest data does not exceed the preset failure threshold of the key performance characteristic parameter, the most recent historical performance characteristic parameter data is selected to form a degradation data time window, and the Monte Carlo simulation method is used to generate the simulated degradation path of the key performance characteristic parameter based on the model parameter value.

[0034] S3. Sampling is performed on the parameter interval in S1 to obtain S simulated degradation paths, and then the failure time of each simulated degradation path is calculated.

[0035] S4. Based on the estimated average failure time and the current time, obtain the remaining service life of the current clutch unit.

[0036] Preferably, the method further includes comparing the prediction performance of the bivariate inverse Gaussian degradation model with that of the bivariate Gamma model, calculating the MAE of the two models respectively, performing quantitative analysis on the prediction performance, and verifying the effectiveness of the bivariate inverse Gaussian degradation model.

[0037] On the other hand, to achieve the above objectives, the present invention also provides an online remaining life prediction system for a dual-variable wet clutch, applied to the online remaining life prediction method for a dual-variable wet clutch, comprising:

[0038] Sample torque storage module: used to acquire real-time friction torque data of the clutch, obtain the friction torque curve, and store it in the torque storage module;

[0039] Torque preprocessing module: used to denoise and smooth the collected friction torque data, eliminate noise and outliers, and extract key performance characteristic parameters based on the smoothed torque data;

[0040] Bivariate Inverse Gaussian Model Module: Constructs a bivariate inverse Gaussian process model based on the processed data;

[0041] Goodness-of-fit test module: used to evaluate the fitting effect of the parameters and paths estimated by the bivariate inverse Gaussian process model;

[0042] RUL Online Prediction Module: Based on the updated model and real-time data, it uses Monte Carlo simulation to sample and predict the future trend of degradation path and dynamically calculate the remaining service life of wet clutches.

[0043] Compared with the prior art, the present invention has the following advantages and technical effects:

[0044] The method of this invention extracts the main performance parameters (friction coefficient and engagement time) based on the real-time friction torque data during clutch operation. By constructing a bivariate inverse Gaussian (IG) degradation model, it accurately captures the correlation between friction plate wear and friction coefficient changes in wet clutches. Combined with Markov chain Monte Carlo (MCMC) and Monte Carlo simulation techniques, it achieves high-precision parameter estimation and real-time remaining life (RUL) prediction.

[0045] The system proposed in this invention not only improves the dynamic response capability of life prediction, but also guides the formulation of condition-based maintenance decisions, reduces the risk of sudden failures, and improves the operational reliability of the system. Attached Figure Description

[0046] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0047] Figure 1 This is a flowchart of an online remaining life prediction method for a dual-variable wet clutch according to an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the structure of an online remaining life prediction system for a dual-variable wet clutch according to an embodiment of the present invention;

[0049] Figure 3 This is a smoothed friction torque diagram according to an embodiment of the present invention;

[0050] Figure 4 This is a diagram showing the evolution of COF and bonding time in an embodiment of the present invention;

[0051] Figure 5 This is a schematic diagram of the online prediction results of RUL in an embodiment of the present invention. Detailed Implementation

[0052] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0053] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0054] The clutch operates primarily by controlling the engagement and disengagement of the friction plates and their mating steel plates to transmit or interrupt power. When the clutch engages, the pressure applied by the hydraulic system forces the piston to push the friction plates into contact with the mating steel plates, transmitting power through friction. When the clutch disengages, the hydraulic pressure is released, the piston returns to its original position, the friction plates separate from the steel plates, and power transmission is interrupted. However, during operation, the accumulated wear of the friction elements gradually affects the clutch's performance. This dynamic change makes clutch degradation somewhat uncertain. To effectively assess the clutch's condition and lifespan, extracting and analyzing key characteristic parameters becomes essential. During clutch operation, the coefficient of friction (COF) determines the clutch's torque transmission capability, while engagement time reflects the clutch's response speed and engagement smoothness. These parameters directly reflect the clutch's operating condition, helping to establish accurate degradation models and providing data support for real-time remaining life prediction and preventative maintenance.

[0055] Traditional bivariate degradation models and online remaining life prediction methods for wet clutches suffer from the following problems: 1. Modeling the multivariate degradation process: The performance degradation of wet clutches is typically characterized by multiple parameters, and these variables are correlated. Existing univariate or simple multivariate models cannot accurately describe the multidimensional degradation behavior of wet clutches. A bivariate model that can effectively capture multiple degradation variables and their correlations is needed; 2. Insufficient parameter estimation accuracy: In the process of wet clutch degradation modeling, existing methods suffer from low accuracy when estimating parameters for complex degradation variables, failing to fully reflect the degradation characteristics under actual working conditions. Efficient numerical methods such as MCMC and Monte Carlo simulations need to be introduced to improve the accuracy of bivariate model parameter estimation; 3. Real-time remaining life prediction: Traditional life prediction methods are usually based on a large amount of historical and life data, which cannot adequately cope with changes in real-time dynamic data, resulting in inaccurate predictions. A model capable of dynamically adjusting using real-time data is needed to accurately predict the remaining useful life (RUL) of wet clutches under actual operating conditions. 4. System reliability and maintenance optimization issues: Existing methods lag behind in reliability assessment and preventative maintenance decision support, making it difficult to effectively schedule preventative maintenance based on accurate life prediction, leading to reduced system reliability and increased maintenance costs. A model capable of providing precise support for decision-making is needed to optimize system health management and maintenance plans, and reduce unexpected failures.

[0056] Based on this, this embodiment proposes an online remaining life prediction method for dual-variable wet clutches, which solves many of the aforementioned problems, such as... Figure 1 ,include:

[0057] Acquire real-time friction torque data of the clutch, and preprocess the real-time friction torque data to extract key performance characteristic parameters;

[0058] A bivariate inverse Gaussian degradation model is constructed based on the key performance characteristic parameters. The clutch RUL is then dynamically predicted online using this model. Specifically, the parameters of the bivariate inverse Gaussian degradation model are estimated using the Markov chain Monte Carlo (MCMC) method, and the clutch RUL is predicted online based on extensive sampling using the Monte Carlo simulation method.

[0059] This embodiment establishes a bivariate degradation process model (including friction coefficient and engagement time) that can capture the performance degradation of the clutch. Combined with real-time online acquisition of clutch characteristic parameters, numerical methods such as Markov chain Monte Carlo (MCMC) and Monte Carlo simulation are used to achieve accurate online RUL prediction of the clutch system during service.

[0060] Further, real-time friction torque data of the clutch is acquired, and the real-time friction torque data is preprocessed, including:

[0061] Real-time friction torque data T(t) is collected by the speed and torque sensor in the clutch system; the collected friction torque data of the engagement process is denoised and smoothed to eliminate noise and outliers.

[0062] Furthermore, the method for extracting the key performance characteristic parameters is as follows:

[0063]

[0064] t e =t b -t a ;

[0065] In the formula, t e s is the engagement time; s is the number of friction pairs; p is the applied pressure; R o and R i These are the outer and inner radii of the friction component, respectively; t a and t d T(t) represents the start and end times of the engagement process, respectively. T(t) is the real-time friction torque, and μ is the friction coefficient COF.

[0066] Specifically, key performance characteristic parameters are extracted using the denoised torque data, including the friction coefficient COF and engagement time.

[0067] Furthermore, a bivariate inverse Gaussian degradation model is constructed, including:

[0068] Performance characteristic parameters are obtained by using friction torque data, including the friction coefficient COF and the degradation process of engagement time. It is assumed that random variables are controlled by probability distributions, and then the probability density function of the degradation process is obtained.

[0069] By setting up n clutch units to conduct degradation tests under the same conditions, each unit i obtains m degradation observations. Given the latent variables of the degradation units, the posterior distribution of the latent variables is obtained, and finally the complete log-likelihood function of the bivariate inverse Gaussian degradation model is obtained.

[0070] Specifically, let the COF and bonding time data be represented as Y1(t) and Y2(t), respectively, where t≥0, and they respectively follow an inverse Gaussian distribution, i.e., Y p (t)~IG(μΛ p (t), η p Λ p (t) 2); where p = 1, 2, representing two degradation processes respectively; where μ is a common latent variable used to describe the dependency between the two degradation processes; Λ(t) is a time-dependent shape function in power-law form. t and q are time and time powers, respectively; η describes the volatility of degradation; therefore, the probability density function (PDF) of the degradation process can be obtained as follows:

[0071]

[0072] In the formula, f(Y) p ) is the probability density function of the corresponding performance characteristic parameter.

[0073] As a standard practice, in this embodiment, it is assumed that the random variable is controlled by a probability distribution. To prevent negative values, it is assumed that μ... -1 TN(ω,κ) follows a truncated normal distribution -2 And it is statistically independent of the parameter η. Furthermore, assume a given μ -1 If the degradation processes of the two performance parameters Y1(t) and Y2(t) are independent of each other, then the degradation process can be represented as Y p (t)|μ -1 ~IG(μΛ) p (t),η p Λ p (t) 2 ), p = 1, 2. Therefore, the PDF of the degradation process can be rewritten as:

[0074] f(Y k (t)|μ -1 )=f(Y p );

[0075] The probability density function of the latent variable is:

[0076]

[0077] In the formula, π(μ) -1 Let φ be the probability density function of the latent variable, ω be the mean parameter, κ be the standard deviation parameter, φ be the probability density function of the standard normal distribution, and Φ be the cumulative distribution function of the standard normal distribution.

[0078] Next, the joint PDF of the two degenerate variables is given as follows:

[0079] f(Y1(t),Y2(t))=∫f(Y1(t),Y2(t),μ -1 )dμ -1 ;

[0080] Where, f(Y1(t),Y2(t),μ -1)=f(Y1(t),Y2(t)μ -1 )π(μ -1 ).

[0081] Assuming n clutch units undergo degradation testing under identical conditions, and each unit i obtains m degradation observations, then the degradation measurements can be expressed as:

[0082]

[0083] Among them, y p (t ij ) is the p-th degenerate variable of unit i at time t. ij The degradation value, while letting y k,i,j =y k (t ij )-y k (t i(j-1) ), β k,i,j =Λ k (t ij )-Λ k (t i(j-1) ).

[0084] Then, given the latent variable μ of the degenerate unit i -1 The likelihood function can be written as:

[0085]

[0086] Then, the latent variable μ can be obtained. -1 The posterior distribution:

[0087]

[0088] Finally, the complete log-likelihood function of the bivariate inverse Gaussian degenerate model is obtained:

[0089]

[0090] Where θ=(ω,κ,η1,η2,q1,q2) is the parameter set of the model to be estimated, and P(·) is the latent variable of the given degenerate unit i. The likelihood function of the model, For m observations of the friction coefficient COF of unit i, For m observations of the engagement time of unit i, Let N be the common latent variable of the i-th clutch unit, and N be the number of clutch units.

[0091] The high dimensionality and multiple parameters make direct optimization of the likelihood function difficult. Therefore, this embodiment will utilize the more practical Bayesian Markov Chain Monte Carlo (MCMC) method to perform posterior parameter inference and uncertainty estimation on the parameter set θ.

[0092] To verify the performance of the bivariate inverse Gaussian degradation model, a goodness-of-fit test is performed. This embodiment provides two test metrics: the Watanabe Akaike Information Criterion (WAIC) and the Mean Absolute Error (MAE), which are calculated as follows:

[0093] WAIC = -2(LPD-k);

[0094]

[0095] In the formula, LPD represents the sum of log-point likelihoods, k represents the effective number of parameters, and Y pre and Y act These are the model-fitted degradation value and the actual degradation value, respectively, where m is the number of predicted points.

[0096] The smaller the values ​​of the two metrics, the better the model performance.

[0097] Furthermore, dynamic online prediction of clutch RUL is performed using a bivariate inverse Gaussian degradation model, including:

[0098] Online prediction of RUL (Residual Limiting Usage) is performed using a validated degradation model. Based on the established degradation model and the concept of first-time failure (TFU), the RUL of the clutch system is iteratively predicted using a time window of degradation data obtained from the performance parameters, based on Monte Carlo simulation. Specifically, through iterative degradation data time windows, the model infers the degradation of the clutch system and estimates parameters based on the latest degradation information, finally obtaining the predicted RUL based on the posterior sample distribution.

[0099] Specifically:

[0100] S1. Input the known historical data of the current time t into the bivariate inverse Gaussian degradation model, and obtain the posterior distribution range of each parameter, including the mean and variance, based on the MCMC method;

[0101] S2. When the latest data does not exceed the preset failure threshold of the key performance characteristic parameter, the most recent l historical performance characteristic parameter data are selected to form a degradation data time window, and the Monte Carlo simulation method is used to generate the simulated degradation path of the key performance characteristic parameter based on the model parameter value.

[0102] S3. Sampling is performed on the parameter interval in S1 to obtain S simulated degradation paths, and then the failure time of each simulated degradation path is calculated.

[0103] S4. Based on the estimated average failure time and the current time, obtain the remaining service life of the current clutch unit.

[0104] Furthermore, the method also includes comparing the prediction performance of the bivariate inverse Gaussian degradation model and the bivariate Gamma model, calculating the MAE of the two models respectively, performing quantitative analysis on the prediction performance, and verifying the effectiveness of the bivariate inverse Gaussian degradation model.

[0105] This embodiment also provides an online remaining life prediction system for dual-variable wet clutches, such as... Figure 2 ,include:

[0106] Sample torque storage module: used to acquire real-time friction torque data of the clutch, obtain the friction torque curve, and store it in the torque storage module;

[0107] Torque preprocessing module: used to denoise and smooth the collected friction torque data, eliminate noise and outliers, and extract key performance characteristic parameters based on the smoothed torque data;

[0108] Bivariate Inverse Gaussian Model Module: Constructs a bivariate inverse Gaussian process model based on the processed data;

[0109] Goodness-of-fit test module: used to evaluate the fitting effect of the parameters and paths estimated by the bivariate inverse Gaussian process model;

[0110] RUL Online Prediction Module: Based on the updated model and real-time data, it uses Monte Carlo simulation to sample and predict the future trend of degradation path and dynamically calculate the remaining service life of wet clutches.

[0111] Furthermore, as another embodiment, the system also includes a performance evaluation module for verifying the effectiveness and performance of the bivariate inverse Gaussian degradation model and the RUL prediction method.

[0112] To more clearly illustrate the technical solution of the present invention, specific embodiments are provided below for description:

[0113] This embodiment provides historical data of a clutch unit to verify the bivariate inverse Gaussian process model and the RUL prediction method.

[0114] Figure 3 and Figure 4 The data evolution diagrams of torque data and two performance characteristics obtained using the sample torque storage module and torque preprocessing module are shown respectively.

[0115] Using a bivariate inverse Gaussian process model, the degradation data of two performance parameters were input into the bivariate inverse Gaussian model (Bi-IG) and the contrastive Gamma model (Bi-Ga), respectively. The parameter estimation results were obtained using the MCMC method. Then, the goodness-of-fit test was performed based on the fitting results, and the results are shown in Table 1.

[0116] Table 1

[0117]

[0118] As can be seen from Table 1, both indices of the proposed inverse Gaussian model are significantly lower than those of the classical Gamma process model, proving its effectiveness.

[0119] Then, using the online RUL prediction module, the clutch system's RUL was estimated in real time based on Monte Carlo simulation, and the error index MAE was calculated. The results are as follows: Figure 5 As shown in Table 2.

[0120] Table 2

[0121]

[0122] The method in this embodiment extracts key performance parameters (friction coefficient and engagement time) based on real-time friction torque data during clutch operation. By constructing a bivariate inverse Gaussian (IG) degradation model, it accurately captures the correlation between friction plate wear and friction coefficient changes in wet clutches. Combined with Markov chain Monte Carlo (MCMC) and Monte Carlo simulation techniques, it achieves high-precision parameter estimation and real-time remaining life (RUL) prediction. The system proposed in this embodiment not only improves the dynamic response capability of life prediction but also guides condition-based maintenance decisions, reduces the risk of sudden failures, and improves system operational reliability.

[0123] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for predicting the remaining life of a dual-variable wet clutch online, characterized in that, include: Acquire real-time friction torque data of the clutch, and preprocess the real-time friction torque data to extract key performance characteristic parameters; A bivariate inverse Gaussian degradation model is constructed based on the key performance characteristic parameters. The clutch RUL is dynamically predicted online using the bivariate inverse Gaussian degradation model. The parameters of the bivariate inverse Gaussian degradation model are estimated using the Markov chain Monte Carlo (MCMC) method. Online prediction of clutch RUL is achieved based on a large number of samples using the Monte Carlo simulation method. Constructing the bivariate inverse Gaussian degradation model includes: Performance characteristic parameters are obtained by using friction torque data, and it is assumed that random variables are controlled by probability distributions to obtain the probability density function of the degradation process. The performance characteristic parameters include the degradation process of the friction coefficient COF and engagement time. By setting up N clutch units to conduct degradation tests under the same conditions, each unit i obtains m degradation observations. Given the latent variables of the degradation units, the posterior distribution of the latent variables is obtained, and finally the complete log-likelihood function of the bivariate inverse Gaussian degradation model is obtained. The probability density function of the degradation process is: Let the extracted friction coefficient COF and bonding time be represented as Y1(t) and Y2(t), respectively, where t≥0, and they both follow an inverse Gaussian distribution, i.e., Y p (t)~IG(μΛ p (t), η p Λ p (t) 2 ); where p = 1, 2, representing two degradation processes respectively; μ is a common latent variable used to describe the dependency between the two degradation processes; Λ(t) is a time-dependent shape function, in power-law form. t and q are time and time powers, respectively; η describes the volatility of degradation, thus yielding the probability density function of the degradation process: In the formula, f(Y) p ) represents the probability density function of the corresponding performance characteristic parameter; Assume latent variable μ -1 TN(ω,κ) follows a truncated normal distribution -2 And statistically independent of the parameter η; then assume a given μ -1 If the degradation processes of the two performance parameters Y1(t) and Y2(t) are independent of each other, then the degradation process can be represented as Y p (t)|μ -1 ~IG(μΛ) p (t), η p Λ p (t) 2 Since p = 1 and 2, the probability density function of the degradation process is: f(Y k (t)|μ -1 )=f(Y p ); The probability density function of the latent variable is: In the formula, π(μ) -1 Let φ be the probability density function of the latent variable, ω be the mean parameter, κ be the standard deviation parameter, φ be the probability density function of the standard normal distribution, and Φ be the cumulative distribution function of the standard normal distribution.

2. The method for predicting the remaining life of a dual-variable wet clutch online according to claim 1, characterized in that, Acquire real-time friction torque data of the clutch, and preprocess the real-time friction torque data, including: Friction torque data is collected in real time by a speed and torque sensor, and the friction torque data is denoised and smoothed to eliminate noise and outliers.

3. The method for predicting the remaining life of a dual-variable wet clutch online according to claim 1, characterized in that, The method for extracting the key performance characteristic parameters is as follows: t e =t b -t a ; In the formula, t e It is the engagement time; s is the number of friction pairs; p is the applied pressure; R o and R i These are the outer and inner radii of the friction component, respectively; t a and t b These represent the start and end times of the engagement process, respectively, where T(t) is the real-time frictional torque, and μ... mean COF is the coefficient of friction.

4. The method for predicting the remaining life of a dual-variable wet clutch online according to claim 1, characterized in that, The complete log-likelihood function of the bivariate inverse Gaussian degenerate model is: In the formula, θ=(ω,κ,η1,η2,q1,q2) is the parameter set of the parameters to be estimated in the model, and P(·) is the likelihood function of the common latent variables of the given clutch unit i. For m observations of the friction coefficient COF of unit i, For m observations of the engagement time of unit i, Let N be the common latent variable of the i-th clutch unit, and N be the number of clutch units.

5. The method for predicting the remaining life of a dual-variable wet clutch online according to claim 1, characterized in that, Dynamic online prediction of clutch RUL using the bivariate inverse Gaussian degradation model includes: The Monte Carlo simulation method is used to perform iterative prediction of the RUL (Restricted Usage Limit) of the clutch system using time windows of the degradation data of the obtained performance parameters. Specifically: S1. Input the known historical data of the current time t into the bivariate inverse Gaussian degradation model, and obtain the posterior distribution range of each parameter, including the mean and variance, based on the MCMC method; S2. When the latest data does not exceed the preset failure threshold of the key performance characteristic parameter, the most recent historical performance characteristic parameter data is selected to form a degradation data time window, and the Monte Carlo simulation method is used to generate the simulated degradation path of the key performance characteristic parameter based on the model parameter value. S3. Sampling is performed on the parameter interval in S1 to obtain S simulated degradation paths, and then the failure time of each simulated degradation path is calculated. S4. Based on the estimated average failure time and the current time, obtain the remaining service life of the current clutch unit.

6. The method for predicting the remaining life of a dual-variable wet clutch online according to claim 1, characterized in that, The method further includes comparing the prediction performance of the bivariate inverse Gaussian degradation model with that of the bivariate Gamma model, calculating the MAE of the two models respectively, performing quantitative analysis on the prediction performance, and verifying the effectiveness of the bivariate inverse Gaussian degradation model.

7. A dual-variable wet clutch online remaining life prediction system, applied to the dual-variable wet clutch online remaining life prediction method according to any one of claims 1-6, characterized in that, include: Sample torque storage module: used to acquire real-time friction torque data of the clutch, obtain the friction torque curve, and store it in the torque storage module; Torque preprocessing module: used to denoise and smooth the collected friction torque data, eliminate noise and outliers, and extract key performance characteristic parameters based on the smoothed torque data; Bivariate Inverse Gaussian Model Module: Constructs a bivariate inverse Gaussian process model based on the processed data; Goodness-of-fit test module: used to evaluate the fitting effect of the parameters and paths estimated by the bivariate inverse Gaussian process model; RUL Online Prediction Module: Based on the updated model and real-time data, it uses Monte Carlo simulation to sample and predict the future trend of degradation path and dynamically calculate the remaining service life of wet clutches.

Images