A wet clutch residual life prediction method and system

Abstract

The application discloses a kind of wet clutch remaining life prediction method and system, comprising: obtaining clutch data to be predicted;The clutch data to be predicted is input to improved IG process model, and the remaining life prediction value of clutch is obtained, the improved IG process model is formed by training set training, and the training set is friction coefficient degradation data.The application helps to prolong the reliable service life of clutch, reduce maintenance cost, and provide effective support for health management and predictive maintenance of vehicle.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of combining equipment reliability engineering with probability and statistical analysis, and in particular relates to a method and system for predicting the remaining life of a wet clutch. Background Technology

[0002] Wet multi-plate clutches are crucial power transmission components in vehicle integrated transmission systems, widely used in various special-purpose vehicles. They primarily rely on the friction pair formed by friction plates and a release plate to achieve power transmission and interruption, thus easily achieving high power transmission and control. However, cumulative damage caused by friction and wear gradually thins and smooths the friction surfaces, causing them to lose their intended design function within their service life. Therefore, effectively monitoring the degradation data of wet clutches, i.e., performance characteristics (PC), provides a reliable historical information basis for effective predictive maintenance measures and prediction of remaining useful life (RUL). Accurate prediction methods can extend the service life and reliability of wet clutches, reducing operating costs during their service life.

[0003] Traditional reliability analysis based on life data incurs higher costs in terms of data acquisition and implementation. Compared to life data, PC more accurately reflects the evolution of clutch performance and includes more life information. 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. Furthermore, failure mechanisms can be used to guide the modeling of product degradation processes, perform reliability analysis, and predict RUL (Reliability Undergoing Range), which is instructive for condition-based maintenance. A suitable degradation model is crucial for representing the degradation characteristics and reliability of clutches. Physics-based methods typically require clear mathematical models to quantitatively characterize the degradation behavior of the system, which is nearly impossible for complex clutch systems. Degradation modeling based on stochastic processes is a more promising option due to its ability to capture the stochastic dynamics and unit heterogeneity (typically including stochastic drift and stochastic vibration) in the degradation process. In particular, inverse Gaussian (IG) processes have stood out in recent research and have proven to be an attractive model for degradation processes.

[0004] While many scholars have achieved some success in exploring the degradation of clutch operating characteristics and predicting remaining service life (RUL) using methods such as finite element simulation and physical models, significant research gaps remain in the degradation analysis and RUL prediction of complex wet clutch systems based on stochastic processes. Many studies have not adequately considered the unit heterogeneity between identical and different clutch devices, including long-term trends and short-term effects. Furthermore, existing research largely relies on large amounts of offline data for model initialization to achieve high model performance, neglecting the importance of online real-time diagnostics, which is unacceptable for expensive clutch systems. Therefore, there is an urgent need for an effective life prediction model based on degradation data to guide the degradation prediction and real-time RUL prediction of clutch systems, influencing condition-based maintenance strategies and health management, and ensuring the efficient and reliable operation of clutches during service. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a method and system for predicting the remaining life of a wet clutch, which can overcome the limitations of existing technologies in clutch degradation analysis and remaining life prediction.

[0006] To achieve the above objectives, the present invention provides a method for predicting the remaining life of a wet clutch, comprising:

[0007] Obtain the clutch data to be predicted;

[0008] The clutch data to be predicted is input into the improved IG process model to obtain the predicted value of the remaining clutch life. The improved IG process model is trained by a training set, which is friction coefficient degradation data.

[0009] The improved IG process model involves introducing random drift and random vibration parameters to modify the parameter structure of the IG process model.

[0010] Optionally, obtaining the friction coefficient degradation data includes:

[0011] Using sensors, data on the initial friction coefficient degradation of the clutch are obtained;

[0012] Outlier removal is performed on the initial friction coefficient degradation data to obtain the friction coefficient degradation data.

[0013] Optionally, the improved IG process model further includes:

[0014] Using the Bayesian method, prior and posterior distributions are defined for the parameters; using the MCMC algorithm, posterior sampling is performed on the IG process model with modified structural parameters to obtain the parameter statistics to be estimated for the improved IG process model.

[0015] Optionally, the random drift parameter is μ, and its probability density function is:

[0016]

[0017] The random vibration parameter is λ, and its probability density function is:

[0018]

[0019] Where ω, κ, η, and γ are the hyperparameters to be estimated in the model, and B is the normalization factor. φ is the gamma function, μ is the random drift parameter, λ is the random oscillation parameter, and φ is the probability density symbol of the standard normal distribution.

[0020] Optionally, the modified IG process model with the structural parameters is as follows:

[0021]

[0022] Where Λ(t) is the time-dependent shape function; y t =Y(t), which is the degenerate observation at time t.

[0023] Optionally, before inputting the clutch data to be predicted into the improved IG process model, the method further includes:

[0024] Obtain the latest degradation observation data;

[0025] Based on the latest degradation observation data, a degradation data with a preset time window length is set.

[0026] Optionally, obtaining the predicted remaining life of the clutch includes:

[0027] Based on the improved IG process model, the degradation state is predicted;

[0028] Determine whether the degradation state exceeds a preset fault threshold, and obtain the predicted value of the remaining life of the clutch based on the determination result.

[0029] The present invention also provides a wet clutch remaining life prediction system, comprising: a bench test module, a data processing module, a numerical modeling module, and a performance prediction module;

[0030] The bench test module is used to design cyclic engagement tests of the clutch system under real vehicle conditions and to collect all COF degradation data of the clutch system from the time it is put into use to the end of the test using sensors.

[0031] The data processing module is used to perform outlier removal processing on the COF degradation data;

[0032] The numerical modeling module is used to construct an improved IG process model with modified parameter structures that include random drift and random vibration.

[0033] The performance prediction module is used to obtain the predicted value of the remaining life of the clutch based on the improved IG process model.

[0034] Compared with the prior art, the present invention has the following advantages and technical effects:

[0035] The model in this invention aims to improve the accuracy and stability of clutch RUL prediction using small-sample degradation data, while reducing reliance on high experimental costs. By effectively analyzing historical clutch degradation data and incorporating random effects, this improved IG process model demonstrates excellent predictive accuracy. This invention helps extend the reliable service life of clutches, reduces maintenance costs, and provides effective support for vehicle health management and predictive maintenance. Attached Figure Description

[0036] 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:

[0037] Figure 1 This is a flowchart of a wet clutch remaining life prediction method according to an embodiment of the present invention;

[0038] Figure 2 This is a comparison chart of the actual and predicted values ​​of the remaining clutch life according to an embodiment of the present invention. Detailed Implementation

[0039] 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.

[0040] 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.

[0041] This invention aims to address the limitations of existing technologies in clutch degradation analysis and remaining life prediction. First, it accurately extracts the coefficient of friction (COF), a key PC parameter characterizing clutch performance degradation, from degradation data. Second, considering the nonlinear characteristics and complex dynamic behavior of clutch systems, an improved IG process model based on small-sample degradation data is developed. By fully utilizing historical data mining and learning degradation characteristics at different time stages, the accuracy and stability of predictions are improved, enabling real-time prediction of clutch RUL (Remaining Life).

[0042] This invention proposes a method for predicting the remaining life of a wet clutch, such as... Figure 1 As shown, it specifically includes:

[0043] Obtain the clutch data to be predicted;

[0044] The clutch data to be predicted is input into the improved IG process model to obtain the predicted value of the remaining clutch life. The improved IG process model is trained using a training set, which consists of friction coefficient degradation data.

[0045] Furthermore, obtaining friction coefficient degradation data includes:

[0046] Using sensors, data on the initial friction coefficient degradation of the clutch are obtained;

[0047] Outlier removal was performed on the initial friction coefficient degradation data to obtain the friction coefficient degradation data.

[0048] Specifically, a cyclic engagement test of the clutch system under real vehicle operating conditions was designed, using sensors to collect all friction coefficient degradation data of the clutch system from its initial use to the end of the test. To control the influence of extraneous variables and ensure the validity of the degradation data, the clutch must be subjected to forced oil cooling after each engagement test.

[0049] The criteria for determining the end of the clutch life cycle test are as follows: the first engagement time of the clutch is t1, and the nth engagement time is tn; if tn ≥ 5 * t1, the life cycle test ends. The life cycle test can also end if smoke (oil vaporization), abnormal noise, or other abnormalities occur during the test.

[0050] The test interval for the clutch cycle test is 100 cycles, meaning that the COF degradation value is collected every 100 engagement cycles.

[0051] Given the inherent nonlinearity and complexity of the clutch system, outliers in the COF degradation data are initially removed and trend smoothing is performed through data preprocessing, which serves as the input PC parameters for the subsequent model.

[0052] Furthermore, the improved IG process model is as follows:

[0053] The parameter structure of the IG process model is modified by introducing random drift parameters and random vibration parameters;

[0054] The Bayesian method was used to test the IG process model with modified structural parameters, and an improved IG process model was obtained.

[0055] Specifically, considering the unit heterogeneity of the clutch system, an improved IG process model IG_RDV with a modified parameter structure incorporating two random effects (random drift and random vibration) is constructed. Subsequently, relevant sub-modules of the model are configured, and prior parameters are set, including the introduction of random effects and their hyperparameters, and the selection of shape functions. Next, sample sampling for posterior prediction testing of the model is implemented using Bayesian methods and Markov chain Monte Carlo (MCMC) algorithms to obtain parameter estimation results and preliminarily verify the model's ability to fit using degenerate data.

[0056] Random drift refers to the uncertainty of the overall trend of a random process over time, and is a long-term characteristic. Random oscillation represents the uncertainty of instantaneous fluctuations in a random process, and is a short-term irregular change superimposed on a long-term trend. Introducing both random drift and random oscillation into the IG process model can better characterize the uncertainty of performance degradation of complex clutch systems in both the long and short-term processes. When using Bayesian MCMC for model parameter inference, the first 800 samples of the simulation process are discarded through a 'burned-in' operation (to stabilize the sampling results within the target distribution region), and the last 1500 samples are used as representatives of the degradation samples.

[0057] The constructed improved IG process model incorporating random effects mainly includes the following steps: For a simple IG process IG(μΛ(t), λΛ(t)... 2 The degradation process Y(t) of μ,λ>0 (abbreviated as IG_Sim), when no random effects are introduced, has constant μ and λ, and its probability density function (PDF) is:

[0058]

[0059] Where y = Y(t) and Λ(t) are time-dependent shape functions, typically linear, power-law, or other forms. Based on the degradation characteristics of COF data, the shape function Λ(t) = t is selected. q .

[0060] When random effects are introduced, the parameter structure of the simple IG process is modified to IG(μΛ(t), μ 3 Λ(t) 2 / λ), and let μ~TN(ω,κ). -2 ), λ~Gamma(η,γ) introduce random drift and random vibration, respectively, where TN and Gamma are the truncated normal and gamma distributions, respectively, and ω,κ,η, and γ are the hyperparameters to be estimated in the model. The PDFs of the random effects parameters μ and λ are as follows:

[0061]

[0062] Where B is the normalization factor, B = 1 - Φ(-κω); It is a gamma function.

[0063] The PDF of the modified IG model IG_RDV is as follows:

[0064]

[0065] The joint posterior distribution of the model can then be expressed as:

[0066] π(ω,κ,η,γ,Λ(t)|Y)∝π(ω,κ,η,γ,Λ(t))L(Yω,κ,η,γ,Λ(t))

[0067] Where π(·) and L(·) are the prior distribution and likelihood function of the parameter to be estimated, respectively.

[0068] Furthermore, the process of inputting the clutch data to be predicted into the improved IG process model also includes:

[0069] Obtain the latest degradation observation data;

[0070] Based on the latest degradation observation data, a degradation data with a preset time window length is set.

[0071] Furthermore, obtaining the predicted remaining life of the clutch includes:

[0072] Based on the improved IG process model, the degradation state is predicted;

[0073] Determine whether the degradation state exceeds the preset fault threshold, and obtain the predicted value of the remaining life of the clutch based on the determination result.

[0074] Specifically, according to the established IG process model and the concept of the first passage time of equipment failure, the remaining useful life (RUL) iterative prediction of the clutch system is realized by using the time window of degradation data (i.e., the preprocessed COF data). That is, through the iterative degradation data time window, the model makes degradation inference and parameter estimation of the clutch system according to the latest degradation information, and finally obtains the predicted RUL based on the posterior sample distribution.

[0075] The steps for RUL prediction using the degradation time window are as follows:

[0076] Let Y i (t 1:k ) = {y i (t1), y i (t2), …, y i (t k )} represent the observed values of the degradation data of the clutch system. According to the preset failure threshold D and the definition of the first passage time, the remaining life of the system at time t is R t = inf{d > 0: Y(t + d) ≥ D|Y(t) < D}. It is difficult to directly obtain the analytical solution of the target distribution, so the Bayesian MCMC method is used to obtain an approximate value of the objective function. The specific steps for RUL calculation are as follows:

[0077] (1) Obtain the latest degradation observation data (COF) at the current time t k , and select the degradation data Y i (t k-l:k ) = {y i (t k-l ), y i (t k-l+1 ), …, y i (t k )} with a preset time window length of l;

[0078] (2) Input the observed data in the time window into the established degradation process model, and obtain the posterior predictive distribution and estimated parameters based on the current information;

[0079] (3) Use the newly estimated parameters and the degradation model to predict the future degradation state, that is, y i (t k + d) = y i (t[[ID=5​​​​​​​​​​​+d) Does it exceed the preset fault threshold D? If not, continue degradation prediction based on the current information; if yes, the fault time and RUL are t and t, respectively. k +d and d.

[0081] Among them, y i (t k +d) represents the predicted future time t k +d's degenerate value; Δy i (t k +d-1) represents the predicted future time t k +d-1 is the degradation increment; d is the degradation model's utilization of the current time t. k The RUL prediction results are obtained from the time window of the degraded data.

[0082] The model's fit performance and complexity are evaluated using the maximum log-likelihood value and the Akaike information criterion (AIC); the model's predictive performance is evaluated using the mean absolute error (MAE).

[0083] The formula for calculating AIC is as follows: Where k represents the number of parameters in the model. This is the maximum log-likelihood value of the model. A smaller AIC indicates better model fit and complexity. Maximum log-likelihood value The likelihood function L is the value at the maximum likelihood estimate, representing the maximum probability or probability density of the observed data (degraded data) under the optimal parameter estimates. To illustrate the goodness of fit of the proposed model, Table 1 presents the maximum log-likelihood and AIC values ​​obtained using COF data from clutch degradation experiments under different IG process models.

[0084] Table 1

[0085]

[0086] This invention uses the Mean Absolute Error (MAE) metric to evaluate the RUL prediction performance of the model, and its expression is as follows:

[0087]

[0088] In the formula, N represents the number of prediction time points from the prediction start time to the actual failure time; D represents the mean of the sampled RUL during prediction; act This is the actual value of RUL.

[0089] The present invention also provides a wet clutch remaining life prediction system, comprising: a bench test module, a data processing module, a numerical modeling module, a performance prediction module, and a data evaluation module;

[0090] The bench test module is used to design cyclic engagement tests of the clutch system under real vehicle conditions, and uses sensors to collect all COF degradation data of the clutch system from the time it is put into use to the end of the test.

[0091] The data processing module performs preliminary data preprocessing to remove outliers from COF-degraded data and smooth the trend.

[0092] The numerical modeling module is used to construct improved IG process models with modified parameter structures that include random drift and random vibration;

[0093] The performance prediction module uses a time window for degraded data to perform iterative prediction of the clutch system's remaining life (RUL) and obtain the predicted value of the clutch's remaining life.

[0094] The data evaluation module assesses the model's fit and complexity based on log-likelihood values ​​and the Akaike Information Criterion, and evaluates the model's predictive performance using mean absolute error (MAE).

[0095] To fully demonstrate the superiority of the RUL prediction model of this invention, the MAE index of the RUL prediction results considering and not considering random effects was calculated on a COF degradation process dataset of a clutch system, as shown in Table 2:

[0096] Table 2

[0097]

[0098]

[0099] To visually represent the error trend between the predicted and actual RUL values, the following is given: Figure 2 The results shown are comparative. It can be seen that this invention can accurately predict the RUL (Recovery Duration and Usage) of the clutch system in real time, thereby enabling timely monitoring of the clutch's working quality and health status, providing a scientific basis for clutch health management and predictive maintenance during service. This technical solution helps extend the clutch's service life, reduce maintenance costs, and ensure the safe and stable operation of the vehicle.

[0100] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the remaining life of a wet clutch, characterized in that, include: Obtain the clutch data to be predicted; The clutch data to be predicted is input into the improved IG process model to obtain the predicted value of the remaining clutch life. The improved IG process model is trained by a training set, which is friction coefficient degradation data. The improved IG process model is achieved by introducing random drift parameters and random vibration parameters to modify the parameter structure of the IG process model. The improved IG process model also includes: Using the Bayesian method, prior and posterior distributions are defined for the parameters; using the MCMC algorithm, posterior sampling is performed on the IG process model with modified structural parameters to obtain the parameter statistics to be estimated for the improved IG process model. The random drift parameters μ The probability density function is: The random vibration parameters λ The probability density function is: in, ω , κ , η and γ These are the hyperparameters to be estimated in the model. B It is a normalization factor. It is the gamma function. μ For random drift parameters, λ These are random vibration parameters. ϕ This is the probability density symbol for the standard normal distribution. The probability density function of the IG process model after the structural parameters are modified is: in, Λ ( t ) is a time-dependent shape function; y t = Y ( t ), which is the degenerate observation at time t.

2. The method for predicting the remaining life of a wet clutch according to claim 1, characterized in that, Obtaining the friction coefficient degradation data includes: Using sensors, data on the initial friction coefficient degradation of the clutch are obtained; Outlier removal is performed on the initial friction coefficient degradation data to obtain the friction coefficient degradation data.

3. The method for predicting the remaining life of a wet clutch according to claim 1, characterized in that, The process of inputting the clutch data to be predicted into the improved IG process model also includes: Obtain the latest degradation observation data; Based on the latest degradation observation data, a degradation data with a preset time window length is set.

4. The method for predicting the remaining life of a wet clutch according to claim 1, characterized in that, Obtaining the predicted remaining life value of the clutch includes: Based on the improved IG process model, the degradation state is predicted; Determine whether the degradation state exceeds a preset fault threshold, and obtain the predicted value of the remaining life of the clutch based on the determination result.

5. A wet clutch remaining life prediction system, characterized in that, include: Bench testing module, data processing module, numerical modeling module, performance prediction module; The bench test module is used to design cyclic engagement tests of the clutch system under real vehicle conditions, and uses sensors to collect all friction coefficient degradation data of the clutch system from the time it is put into use to the end of the test. The data processing module is used to perform outlier removal processing on the friction coefficient degradation data; The numerical modeling module is used to construct an improved IG process model with modified parameter structures that include random drift and random vibration. The performance prediction module is used to obtain the predicted value of the remaining life of the clutch based on the improved IG process model. The improved IG process model also includes: Using the Bayesian method, prior and posterior distributions are defined for the parameters; using the MCMC algorithm, posterior sampling is performed on the IG process model with modified structural parameters to obtain the parameter statistics to be estimated for the improved IG process model. Random drift parameters μ The probability density function is: Random vibration parameters λ The probability density function is: in, ω , κ , η and γ These are the hyperparameters to be estimated in the model. B It is a normalization factor. It is the gamma function. μ For random drift parameters, λ These are random vibration parameters. ϕ This is the probability density symbol for the standard normal distribution. The probability density function of the IG process model after the structural parameters are modified is: in, Λ ( t ) is a time-dependent shape function; y t = Y ( t ), which is the degenerate observation at time t.

Images