Diesel engine service life prediction method based on support vector machine model

Abstract

The invention is suitable for the technical field of diesel engines, and provides a diesel engine service life prediction method based on a support vector machine model, which comprises the following steps of: extracting state parameters from full-life test data of a known diesel engine of a certain model; constructing a support vector machine model, setting input and output variables, and training the support vector machine model; carrying out extended Kalman filtering on data from the support vector machine model, setting initial parameters, and carrying out iterative calculation; and establishing a function relationship between the state parameters and the residual life according to data calculated by the training sample, and obtaining an estimation interval of the residual life of the engine. The invention provides a life prediction model combining a support vector machine and an extended Kalman filtering method, through theoretical result comparison and simulation example verification, data errors can be greatly corrected, the operand is reduced, stable and accurate prediction of the residual life of the engine is realized, and the service life of the engine is prolonged. And a reliable vehicle maintenance reference basis is provided for a vehicle equipped with a diesel engine.

Description

Technical Field

[0001] The invention belongs to the technical field of diesel engines, and in particular relates to a diesel engine life prediction method based on a support vector machine model. Background Art

[0002] Predicting the life of diesel engines is crucial for heavy-duty work vehicles, as these vehicles are not only costly but also carry out critical transportation and production tasks. By timely predicting the remaining life of the engine, maintenance and care can be effectively arranged to prevent failures, ensure driving safety and smooth completion of tasks. In addition, reasonable arrangements for inspections and maintenance based on the results of diesel engine life prediction can also avoid the waste of manpower and material resources caused by excessive irregular inspections, thereby maximizing the balance between safety and economy.

[0003] In the structure of diesel engines, the cylinder liner is a key component that determines the life of the engine. Its wear degree is currently considered to be the main basis for evaluating whether the engine has failed. Due to the working characteristics of reciprocating piston diesel engines, as the vehicle mileage increases, the cylinder liner will produce uneven wear, resulting in cylinder leakage, cylinder pressure drop, and ultimately the end of the engine life. There are currently two main methods for predicting the life of diesel engines: The first is the rated life prediction method, which assumes that the same model of engine has a unified "rated life" or maximum life value. When calculating the remaining life of the engine, first read the mileage that the vehicle has traveled, and then use regression prediction, statistical prediction, and empirical prediction data processing models to predict the rated life. However, this method has great limitations. Due to the different environmental conditions of vehicle driving, the degree of wear and aging of the engine varies under the same mileage. The rated life calculated using only the single variable of mileage cannot reflect the individual differences of the engine. As a result, some engines in better condition will be overhauled in advance, causing huge economic losses. At the same time, some engines in poorer condition will not be overhauled in time, causing certain safety hazards.

[0004] The second is to read the three characteristic parameters of cylinder pressure, injection pressure and oil pressure in real time, and use the empirical formula with correction coefficient to calculate the remaining life of the engine. This method can dynamically predict the life in real time, but because there is no feedback adjustment mechanism in its mathematical model to eliminate errors, the accuracy of its prediction is insufficient, ranging from 2% to 3%, and when the sensor is subject to greater external interference, the accuracy of the remaining life prediction result will become unstable. Therefore, a diesel engine life prediction method based on support vector machine model is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a diesel engine life prediction method based on a support vector machine model, aiming to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: A diesel engine life prediction method based on a support vector machine model comprises the following steps: Step 1, extracting state parameters from the full life test data of a known diesel engine model; Step 2: construct a support vector machine model, set input and output variables, and train the support vector machine model; Step 3: Perform extended Kalman filtering on the data from the support vector machine model, set initial parameters, and iterate the calculation; Step 4: Establish a functional relationship between the state parameters and the remaining life based on the data calculated from the training samples to obtain an estimated interval of the remaining life of the engine.

[0007] Furthermore, the specific process of step 1 is as follows: Three characteristic state parameters for measuring the degree of engine wear are extracted from the full life test data of a known diesel engine model, including cylinder pressure, injection pressure, oil pressure and remaining service life. The three characteristic state parameters are used as training samples to establish a support vector machine model.

[0008] Furthermore, the specific process of step 2 is as follows: Construct a support vector machine model and input variables x The cylinder pressure, injection pressure and oil pressure are set respectively, and the output variable is set to the remaining life of the engine. Then, the optimal parameter combination is determined through practical experience, and the support vector machine model is trained to finally obtain the regression function between the output variable and the input variable. : ; ; in N is 3; is the nth parameter estimate; is the updated value of the nth parameter estimate; is the kernel function of the linear regression function; is the intercept estimate of the regression function; is the empirical coefficient, ranging from 0 to 1; is the characteristic parameter of the nth input.

[0009] Furthermore, the specific process of step 3 is as follows: Perform extended Kalman filtering on the data output by the support vector machine model. The steps are as follows: First initialize the Kalman filter: ; ; in is the parameter value at time 0; is the parameter estimate at time 0; E is the mathematical expectation of the matrix; is the covariance matrix at time 0; The parameter estimate for the next moment after time 0; For the matrix The transpose of T is the transpose operation; Then set the initial remaining life point estimate and variance, start iterative calculation, and use k -1 time state value estimation k Status value at the moment: ; in for k The time before the moment Estimation of measured values; for k- 1 moment ago Estimation of measured values; is the observed noise value, which is generally assumed to obey Gaussian distribution; is the state transfer function, where the parameter values ​​A and B can determine their optimal values ​​according to the input data; Calculation using the Jacobian matrix of the state equation k The forecast covariance matrix at time : ; in is the Jacobian matrix of the input data; for k -1 The covariance of the forecast error at time is the covariance matrix of the observation noise; calculate k The Kalman gain matrix required for state estimation at the moment : ; in is the measurement matrix; Update with measurements k State estimation at time: ; in for kMoment parameter estimates; is the real-time measurement value of the status; h The state value of the passed model; Using the Kalman gain matrix and k The covariance matrix of the time points before the moment renew k The covariance matrix at time : ; in, I is the identity matrix.

[0010] Furthermore, the specific process of step 4 is as follows: A life prediction model is established based on the data obtained by iterative calculation of training samples, and the functional relationship between the engine state parameters and the remaining life is fitted to establish the functional relationship between the state parameters and the remaining life; the threshold for the end of engine life is set according to the maximum cylinder leakage or minimum cylinder pressure specified in the standard. When the engine state degrades to the effective threshold, the engine life is considered to have ended. The maximum likelihood estimation method is used for estimation, and the functional relationship between the state parameters and the remaining life is used to obtain the estimated interval of the engine's remaining life.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The present invention has practical significance in the actual application to the life prediction of diesel engines. The life prediction model combining support vector machine and extended Kalman filter method has been verified through theoretical result comparison and simulation example verification, and can indeed greatly correct data errors, reduce the amount of calculation, and achieve stable and accurate prediction of the remaining life of the engine, providing a reliable reference for vehicle maintenance for vehicles equipped with diesel engines. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the present invention.

[0013] Figure 2 It is the training set iterative residual in Example 1 of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0015] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0016] like Figure 1As shown, a diesel engine life prediction method based on a support vector machine model provided by an embodiment of the present invention includes the following steps: Step 1: extracting state parameters from the full life test data of a known diesel engine of a certain model; the specific process of step 1 is as follows: Three characteristic state parameters for measuring the degree of engine wear are extracted from the full life test data of a known diesel engine model, including cylinder pressure, injection pressure, oil pressure and remaining service life. Subsequently, the three characteristic state parameters are used as training samples to establish a support vector machine model.

[0017] Step 2: construct a support vector machine model, set input and output variables, and train the support vector machine model; the specific process of step 2 is as follows: Construct a support vector machine model and input variables x The cylinder pressure, injection pressure and oil pressure are set respectively, and the output variable is set to the remaining life of the engine. Then, the optimal parameter combination is determined through practical experience, and the support vector machine model is trained to finally obtain the regression function between the output variable and the input variable. : ; ; in is the nth parameter estimate; is the updated value of the nth parameter estimate; is the kernel function of the linear regression function; is the intercept estimate of the regression function; is the empirical coefficient, with a value of 0~1; is the characteristic parameter of the nth input; in this embodiment N is 3.

[0018] Step 3: Perform extended Kalman filtering on the data from the support vector machine model, set initial parameters, and iterate the calculation; the specific process of step 3 is as follows: Perform extended Kalman filtering on the data output by the support vector machine model. The steps are as follows: First initialize the Kalman filter: ; ; in is the parameter value at time 0; is the parameter estimate at time 0; E is the mathematical expectation of the matrix; is the covariance matrix at time 0; The parameter estimate for the next moment after time 0; For the matrix The transpose of T is the transpose operation; Then set the initial remaining life point estimate and variance, start iterative calculation, and use k -1 time state value estimation k Status value at the moment: ; in for k The time before the moment Estimation of measured values; for k- 1 moment ago Estimation of measured values; is the observed noise value, which is generally assumed to obey Gaussian distribution; is the state transfer function, where the parameter values ​​A and B can determine their optimal values ​​according to the input data; Calculation using the Jacobian matrix of the state equation k The forecast covariance matrix at time : ; in is the Jacobian matrix of the input data; for k -1 The covariance of the forecast error at time is the covariance matrix of the observation noise; calculate k The Kalman gain matrix required for state estimation at the moment : ; in is the measurement matrix; Update with measurements k State estimation at time: ; in for k Moment parameter estimates; is the real-time measurement value of the status; h Is the state value of the passed model; Using the Kalman gain matrix and k The covariance matrix of the time points before the moment renew k The covariance matrix at time : ; in,I is the identity matrix.

[0019] Step 4: Establish a functional relationship between the state parameter and the remaining life according to the data calculated from the training sample to obtain an estimated interval of the remaining life of the engine; the specific process of step 4 is as follows: A life prediction model is established based on the data obtained by iterative calculation of training samples, and the functional relationship between the engine state parameters and the remaining life is fitted to establish the functional relationship between the state parameters and the remaining life; the threshold for the end of engine life is set according to the maximum cylinder leakage or minimum cylinder pressure specified in the standard. When the engine state degrades to the effective threshold, the engine life is considered to have ended, and the maximum likelihood estimation method is used for estimation. The estimated interval of the remaining life of the engine is obtained using the above functional relationship.

[0020] In the embodiment of the present invention, the support vector machine is a machine learning algorithm based on statistical theory. The basic idea is to solve a separating hyperplane so as to correctly divide the training data set and maximize its geometric interval. The characteristic of the support vector machine is that under the condition of limited learning samples, it can obtain the sample state prediction value with good stability and high accuracy.

[0021] Kalman filtering, as a filtering method, is essentially a process of signal processing and transformation. It is a data processing technology that removes data bias and restores real data. It can estimate the state of a dynamic system from a series of data with measurement noise and achieve the optimal estimation of the system state. The basic process of Kalman filtering theory is: establish a state space model and an observation model, estimate the state quantity at the current moment based on the estimated value at the previous moment and the observed value at the current moment, and recursively use the process of "prediction-measurement-correction" to eliminate random interference based on the system's measured value and restore its true state from the contaminated system.

[0022] The extended Kalman filter method is an extension of the application of the Kalman filter method to nonlinear systems. In practical applications, diesel engines are often in a complex and changeable working environment, and their state detection sensors are greatly disturbed by vibration, temperature and noise. The state parameters of the engine often show nonlinear correlation characteristics. In order to continue to apply the Kalman filter algorithm in such a nonlinear system, it is necessary to linearize the nonlinear state model and then apply the Kalman filter method to achieve recursive estimation. This method is the extended Kalman filter method.

[0023] Example 1. In order to verify the effectiveness of the method proposed in the present invention, we use the system simulation software Matlab to build a real-time prediction model for diesel engines that combines support vector machines and extended Kalman filtering methods, and verify it through a diesel engine life simulation data set released by the China North Engine Research Institute. The data set has different simulation parameters and working conditions and has been widely used in the field of engine remaining life prediction. The data set is divided into a training set and a test set. The training set is divided into seven subsets, recording the sampling values ​​of each time point in the complete cycle of the engine from the start of operation to the occurrence of a failure, including the degradation data of the engine; and the test set is divided into four subsets, recording the state parameter values ​​at a certain time point before the engine fails, including the failure data of the engine. The purpose of this simulation is to predict the remaining service life of the engine in the test set and compare it with the training set to verify the feasibility of the method.

[0024] The test set evaluation index values ​​are shown in Table 1:

[0025] See also Figure 1 From the training results of the simulation model, it can be seen that with the increase of the number of iterations, the output results of the training set have good convergence, and the iterative residuals are stable at 1×10 -2 (1e-02) and below. As shown in Table 1, the test results of the test set all have good fitting. Among the four commonly used evaluation indicators of machine learning, RMSE, MAE, and MAPE are all less than 2.5, and R 2 The values ​​are greater than 0.97, indicating that the model can fit the original data well and has good generalization ability.

[0026] In order to better illustrate the effectiveness of the method proposed in the present invention, we designed an engine bench test, and measured the experimental average values ​​of the remaining life of the three engines and the engine state parameters during the test by testing the failure of three engines, and compared them with the predicted values ​​to obtain the relative prediction error of the method. The results are shown in Table 2:

[0027] From Table 2, we can see that the average error of this method in predicting engine life is within 1%, which proves the feasibility and effectiveness of this method.

[0028] In summary, the life prediction model combining the support vector machine and the extended Kalman filter proposed in the present invention is of great help for the decision-making of early maintenance of the engine. The predicted value of the remaining life of the engine predicted by the method proposed in the present invention is closer to the true value and has a smaller error. Whether compared with the traditional single variable rated prediction method or the multi-input dynamic prediction model, the method proposed in the present invention performs better in multiple measurement indicators. This shows that the life prediction model involved in the present invention can fully extract the engine fault degradation characteristics and accurately capture the relevant features required for early prediction, and has strong prediction capabilities and high feasibility and effectiveness.

[0029] The above are only preferred embodiments of the present invention. It should be pointed out that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention. These should also be regarded as the protection scope of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent.

Claims

1. A diesel engine life prediction method based on a support vector machine model, characterized in that: The following steps are involved: Step 1, extracting state parameters from the full life test data of a known diesel engine model; Step 2: construct a support vector machine model, set input and output variables, and train the support vector machine model; Step 3: Perform extended Kalman filtering on the data from the support vector machine model, set initial parameters, and iterate the calculation; Step 4: Establish a functional relationship between the state parameters and the remaining life based on the data calculated from the training samples to obtain an estimated interval of the remaining life of the engine.

2. The diesel engine life prediction method based on the support vector machine model according to claim 1 is characterized in that: The specific process of step 1 is as follows: Three characteristic state parameters for measuring the degree of engine wear are extracted from the full life test data of a known diesel engine model, including cylinder pressure, injection pressure, oil pressure and remaining service life. The three characteristic state parameters are used as training samples to establish a support vector machine model.

3. The diesel engine life prediction method based on the support vector machine model according to claim 2 is characterized in that: The specific process of step 2 is as follows: Construct a support vector machine model and input variables x The cylinder pressure, injection pressure and oil pressure are set respectively, and the output variable is set to the remaining life of the engine. Then, the optimal parameter combination is determined through practical experience, and the support vector machine model is trained to finally obtain the regression function between the output variable and the input variable. : ； ； in N is 3; is the nth parameter estimate; is the updated value of the nth parameter estimate; is the kernel function of the regression function; is the intercept estimate of the regression function; is the empirical coefficient, ranging from 0 to 1; is the characteristic parameter of the nth input.

4. The diesel engine life prediction method based on the support vector machine model according to claim 3 is characterized in that: The specific process of step 3 is as follows: Perform extended Kalman filtering on the data output by the support vector machine model. The steps are as follows: First initialize the Kalman filter: ； ； in is the parameter value at time 0; is the parameter estimate at time 0; is the mathematical expectation of the matrix; is the covariance matrix at time 0; The parameter estimate for the next moment after time 0; For the matrix The transpose of T is the transpose operation; Then set the initial remaining life point estimate and variance, start iterative calculation, and use the state transfer function to k -1 time state value estimation k Status value at the moment: ； in for k The time before the moment Estimation of measured values; for k- 1 moment ago Estimation of measured values; is the observed noise value, which is assumed to obey Gaussian distribution; is the state transfer function, where the parameter values ​​A and B are determined according to the input data; Calculation using the Jacobian matrix of the state equation k The forecast covariance matrix at time : ； in is the Jacobian matrix of the input data; for k -1 The covariance of the forecast error at time is the covariance matrix of the observation noise; calculate k The Kalman gain matrix required for state estimation at the moment : ； in is the measurement matrix; Update with measurements k State estimation at time: ； in for k Moment parameter estimates; is the real-time measurement value of the status; h The state value of the passed model; Using the Kalman gain matrix and k The covariance matrix of the time points before the moment renew k The covariance matrix at time : ； in, I is the identity matrix.

5. The diesel engine life prediction method based on the support vector machine model according to claim 4 is characterized in that: The specific process of step 4 is as follows: A life prediction model is established based on the data obtained by iterative calculation of training samples, and the functional relationship between the engine state parameters and the remaining life is fitted to establish the functional relationship between the state parameters and the remaining life; the threshold for the end of engine life is set according to the maximum cylinder leakage or minimum cylinder pressure specified in the standard. When the engine state degrades to the effective threshold, the engine life is considered to have ended. The maximum likelihood estimation method is used for estimation, and the functional relationship between the state parameters and the remaining life is used to obtain the estimated interval of the engine's remaining life.

Images