A fatigue life prediction method for automotive turbocharger pipelines

Abstract

The invention discloses a fatigue life prediction method for an automobile turbocharger pipeline. The method is characterized in that the S-N curve of the automobile turbocharger pipeline is first measured and obtained; a simulation calculation model of the pipeline is constructed, and the fatigue life value of the pipeline under different conditions is calculated by the simulation calculation model to obtain a data set; an MLP neural network prediction model is constructed, and the MLP neural network prediction model is optimized by a Bayesian optimization algorithm to seek an optimal hyperparameter combination; the data set is substituted into the MLP neural network prediction model with the optimal hyperparameter combination for training; the fatigue life value of the turbocharger pipeline is predicted by the trained MLP neural network prediction model; the method has the advantage that the data samples calculated by the simulation calculation model under different structural parameters and external loads have higher credibility and accuracy, and the MLP neural network prediction model is parameter-tuned by a Bayesian optimization algorithm, so that the prediction result has higher credibility and accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of life prediction of automobile parts, and in particular to a fatigue life prediction method for automobile turbocharger pipelines. Background Art

[0002] With the rapid development of the automobile industry, the performance and safety of automobiles have received extensive attention. As one of the main components of automobile engines, the turbocharger pipeline of automobiles is affected by the vibration of the engine, the pressure of the liquid on the inner wall of the pipeline, and the excitation caused by the uneven road surface during the driving process of the automobile. As a result, the turbocharger pipeline works in such a harsh environment for a long time, which accelerates the fatigue failure of the pipeline and produces cracks, which directly affects the performance and reliability of the engine, and may even cause the turbocharger pipeline to break, causing safety accidents. Therefore, the prediction of the life of the automobile turbocharger pipeline is not only an engineering and technical issue, but also an important guarantee for the performance, safety and reliability of the automobile.

[0003] At present, the fatigue life prediction of automotive turbocharger pipelines basically requires a large amount of experimental data and computing resources to build models and verify prediction results. However, these data are not easy to obtain. For example, pipeline durability testing and long-term monitoring in actual operation require a lot of time and are expensive, which limits the practicality and promotion of prediction methods, making it impossible for engineers to obtain the fatigue life of automotive turbocharger pipelines under different conditions in a short time. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a fatigue life prediction method for an automobile turbocharger pipeline, which can quickly and effectively evaluate the fatigue life of the turbocharger pipeline under different operating conditions, and the reliability and accuracy of the prediction results are high.

[0005] The technical solution adopted by the present invention to solve the above technical problem is: a fatigue life prediction method for automobile turbocharger pipeline, comprising the following specific steps:

[0006] (1) Measure and obtain the SN curve of the automobile turbocharger pipeline;

[0007] (2) Construct a simulation model of the turbocharger pipeline based on the measured SN curve, and compare the fatigue life results obtained by the simulation model with the fatigue life results obtained by the automobile turbocharger pipeline on the vibration test bench to verify the accuracy of the simulation model;

[0008] (3) Analyze the factors that have a great impact on the fatigue life of the turbocharger pipeline based on the simulation calculation model and use them as the feature input of the subsequent prediction model. Then, make different combinations of the factors with the greatest impact in the analysis results to obtain the condition set of the turbocharger pipeline. Calculate the fatigue life value of the turbocharger pipeline under each condition through the simulation calculation model, and integrate the calculated fatigue life data with the fatigue life data obtained by the vibration table to obtain a data set.

[0009] (4) Construct an MLP neural network prediction model, taking the influencing factors in the integrated data set as feature input variables and the fatigue life cycle as output;

[0010] (5) Optimize the MLP neural network prediction model through the Bayesian optimization algorithm. Substitute the data set integrated in the above step (3) into the MLP neural network prediction model optimized by the Bayesian algorithm for training to find the optimal hyperparameter combination;

[0011] (6) Substitute the data set integrated in step (3) above into the MLP neural network prediction model with the optimal hyperparameter combination for training;

[0012] (7) The fatigue life value of the turbocharger pipeline is predicted by the MLP neural network prediction model trained in step (6).

[0013] Furthermore, the specific method for obtaining the SN curve in step (1) is as follows:

[0014] (1-1) Assemble the automobile turbocharger pipeline on a tensile test bench and perform tensile tests under different displacements to obtain the number of fatigue cycle of the turbocharger pipeline under different displacements;

[0015] (1-2) Calculate the stress amplitude of the turbocharger pipeline under different displacements through simulation;

[0016] (1-3) The fatigue cycle number under different displacements obtained from the tensile test is fitted with the corresponding stress amplitude obtained from the simulation to obtain the SN curve, and the relationship is: , where: S represents the stress amplitude of the turbocharger pipeline; N Indicates the number of fatigue cycles of the turbocharger pipeline; m, C They are the fatigue strength index and fatigue strength coefficient of the turbocharger pipeline respectively.

[0017] Furthermore, in the step (3), the specific method for obtaining the factors that have a great influence on the fatigue life of the turbocharger pipeline is: based on the simulation calculation model, a control variable method is used to study the influence of the structural parameters of the pipeline and the external load factors on the fatigue life of the pipeline, and the factors that have a great influence on the fatigue life of the turbocharger pipeline are analyzed.

[0018] Furthermore, in the step (3), the condition set of the turbocharger pipeline is obtained by implementing an optimal Latin hypercube sampling method.

[0019] Furthermore, the specific process of step (5) is as follows:

[0020] (5-1) Divide the data set integrated in step (3) into a training set and a test set;

[0021] (5-2) The MLP neural network prediction model is used as the objective function to be optimized, and the hyperparameters: the number of hidden layers of the MLP neural network, the number of neurons in each hidden layer, the learning rate, the learning decay coefficient, the number of training times, and the regularization coefficient are used as independent variables. The average relative error between the fatigue life value of the pipeline predicted by the prediction model for the test set and the fatigue life value of the pipeline in the test set is used as the target result;

[0022] (5-3) Randomly select several groups of hyperparameters as initial sample points, obtain their corresponding function values ​​from the objective function, and use Gaussian process to fit the proxy probability function model, establish the probability distribution of the function, and obtain the expected mean and variance of each hyperparameter group, where: the mean represents the final expected effect of the hyperparameter group, and the variance represents the uncertainty of the effect of the hyperparameter group. The larger the variance, the greater the uncertainty of the effect of the hyperparameter group.

[0023] (5-4) Based on the above fitted proxy probability function, further accurately fit the proxy probability function, and gradually approach the target function by sampling in the area with large variance, so that the proxy probability function continuously approaches the target function;

[0024] (5-5), repeat steps (5-3) and (5-4) until the stopping condition is reached;

[0025] (5-6) The optimal hyperparameter combination is obtained by fitting the proxy probability function.

[0026] Furthermore, the prediction method of step (7) is as follows: inputting the structural parameters and external loads of the new turbocharger pipeline model into the MLP neural network prediction model to calculate the fatigue life value of the turbocharger pipeline.

[0027] Compared with the prior art, the advantages of the present invention are:

[0028] (1) This method starts from the SN curve actually measured on the turbocharger pipeline and constructs a simulation calculation model based on the data measured on the real molded pipeline, so that the data samples calculated by the simulation calculation model under different structural parameters and external loads have high credibility and accuracy.

[0029] (2) This method uses the Bayesian optimization algorithm to tune the parameters of the MLP neural network prediction model. The Bayesian algorithm adjusts the search space according to the performance of the parameters that have been searched in the past, focusing on the parameter combinations with higher probability, improving the search efficiency; and it can consider the dependencies between different hyperparameters, so as to search and adjust the parameters more intelligently. It improves the problem of overly random selection of MLP neural network hyperparameters, and can provide a more efficient and intelligent search strategy, so that the MLP neural network prediction model can quickly and effectively evaluate the fatigue life of the turbocharger pipeline under different operating conditions, and the prediction results are more reliable and accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of the method of the present invention;

[0031] Figure 2 The figure is a training result diagram of the MLP neural network prediction model using the Bayesian optimization algorithm of the present invention;

[0032] Figure 3 The training result diagram of the MLP neural network prediction model with the training set in the data set in step (3) substituted into the optimal hyperparameter combination;

[0033] Figure 4 The figure is a training result diagram of the MLP neural network prediction model that substitutes the test set in the data set in step (3) into the optimal hyperparameter combination. DETAILED DESCRIPTION

[0034] The present invention is further described in detail below with reference to the accompanying drawings.

[0035] like Figure 1 As shown, a fatigue life prediction method for a pipeline of an automobile turbocharger comprises the following specific steps:

[0036] (1) Measure and obtain the SN curve of the automotive turbocharger pipeline, specifically:

[0037] (1-1) Assemble the automotive turbocharger pipeline on a tensile test bench and conduct tensile tests under different displacements to obtain the number of fatigue cycles of the turbocharger pipeline under different displacements; and in order to prevent unrecognizable fatigue damage, perform air tightness detection on the pipeline after every 3,000 cycles;

[0038] (1-2) Calculate the stress amplitude of the turbocharger pipeline under different displacements through simulation;

[0039] (1-3) The fatigue cycle number under different displacements obtained from the tensile test is fitted with the corresponding stress amplitude obtained from the simulation to obtain the SN curve, and the relationship is: , where: S represents the stress amplitude of the turbocharger pipeline; N Indicates the number of fatigue cycles of the turbocharger pipeline; m, C They are respectively the fatigue strength index and fatigue strength coefficient of the turbocharger pipeline;

[0040] (2) Construct a simulation model of the turbocharger pipeline based on the measured SN curve, and compare the fatigue life results obtained by the simulation model with the fatigue life results obtained by the automobile turbocharger pipeline on the vibration test bench to verify the accuracy of the simulation model; if the error between the two is large, the simulation model is gradually corrected and optimized by existing methods until the error between the two is within the allowable range;

[0041] (3) Based on the simulation calculation model, the control variable method is used to study the influence of the structural parameters of the pipeline (such as the wall thickness of the corrugated pipeline, the wave height of the corrugated pipeline, the wave distance of the corrugated pipeline, etc.) and the external load factors (such as the liquid pressure, the load size, etc.) on the fatigue life of the pipeline. The factors that have a great influence on the fatigue life of the turbocharger pipeline are analyzed and used as the feature input of the subsequent prediction model. Then, by implementing the optimal Latin hypercube sampling method, different combinations of the factors with the greatest influence in the analysis results are made to obtain the condition set of the turbocharger pipeline. The fatigue life value of the turbocharger pipeline under each condition is calculated by the simulation calculation model, and the calculated fatigue life data is integrated with the fatigue life data obtained by the vibration table to obtain a data set;

[0042] (4) Use Python to build an MLP neural network prediction model, taking the influencing factors in the integrated data set as feature input variables and the fatigue life cycle as output;

[0043] (5) The MLP neural network prediction model is optimized by the Bayesian optimization algorithm to find the optimal hyperparameter combination, specifically:

[0044] (5-1) Divide the data set integrated in step (3) into a training set and a test set;

[0045] (5-2) The MLP neural network prediction model is used as the objective function to be optimized, and the hyperparameters: the number of hidden layers of the MLP neural network, the number of neurons in each hidden layer, the learning rate, the learning decay coefficient, the number of training times, and the regularization coefficient are used as independent variables. The average relative error between the fatigue life value of the pipeline predicted by the prediction model for the test set and the fatigue life value of the pipeline in the test set is used as the target result, such as Figure 2 As shown;

[0046] (5-3) Randomly select several sets of hyperparameters as initial sample points, obtain their corresponding function values ​​from the objective function, and use Gaussian process Fit the proxy probability function model, establish the probability distribution of the function, and obtain the expected mean and variance of each hyperparameter group, where: is the mean function, which is used to obtain the expected mean of each hyperparameter group. is the covariance function, which is used to obtain the variance of each hyperparameter group; the mean represents the final expected effect of the hyperparameter group, and the variance represents the uncertainty of the effect of the hyperparameter group. The larger the variance, the greater the uncertainty of the effect of the hyperparameter group.

[0047] (5-4) Based on the above fitted proxy probability function, further accurately fit the proxy probability function, sample in the area with large variance (usually refers to the area with high model prediction uncertainty) through the expected improvement method (i.e., expected improvement function), and gradually approach the target function, so that the proxy probability function continuously approaches the target function;

[0048] (5-5), repeat steps (5-3) and (5-4) until the stopping condition is reached;

[0049] (5-6) Obtain the optimal hyperparameter combination through the fitted proxy probability function;

[0050] (6) After the data set integrated in step (3) is standardized, it is put into the MLP neural network prediction model with the optimal hyperparameter combination for training. The training results are as follows: Figure 3 , 4 As shown;

[0051] (7) Input the structural parameters and external loads of the new turbocharger pipeline model into the MLP neural network prediction model trained in step (6) to calculate the fatigue life value of the turbocharger pipeline.

[0052] The protection scope of the present invention includes but is not limited to the above embodiments, and its protection scope is subject to the claims. Any replacement, deformation, and improvement of the technology that can be easily thought of by technicians in this field shall fall within the protection scope of the present invention.

Claims

1. A fatigue life prediction method for automobile turbocharger pipeline, characterized in that The specific steps include: (1) Measure and obtain the SN curve of the automobile turbocharger pipeline; (2) Construct a simulation model of the turbocharger pipeline based on the measured SN curve, and compare the fatigue life results obtained by the simulation model with the fatigue life results obtained by the automobile turbocharger pipeline on the vibration test bench to verify the accuracy of the simulation model; (3) Analyze the factors that have a great impact on the fatigue life of the turbocharger pipeline based on the simulation calculation model and use them as the feature input of the subsequent prediction model. Then, make different combinations of the factors with the greatest impact in the analysis results to obtain the condition set of the turbocharger pipeline. Calculate the fatigue life value of the turbocharger pipeline under each condition through the simulation calculation model, and integrate the calculated fatigue life data with the fatigue life data obtained by the vibration table to obtain a data set. (4) Construct an MLP neural network prediction model, taking the influencing factors in the integrated data set as feature input variables and the fatigue life cycle as output; (5) The MLP neural network prediction model is optimized by using the Bayesian optimization algorithm. The data set integrated in the above step (3) is substituted into the MLP neural network prediction model optimized by the Bayesian algorithm for training to find the optimal hyperparameter combination; specifically: (5-1) Divide the data set integrated in step (3) into a training set and a test set; (5-2) The MLP neural network prediction model is used as the objective function to be optimized, and the hyperparameters: the number of hidden layers of the MLP neural network, the number of neurons in each hidden layer, the learning rate, the learning decay coefficient, the number of training times, and the regularization coefficient are used as independent variables. The average relative error between the fatigue life value of the pipeline predicted by the prediction model for the test set and the fatigue life value of the pipeline in the test set is used as the target result; (5-3) Randomly select several sets of hyperparameter groups as initial sample points, obtain their corresponding function values ​​from the objective function, and use Gaussian process to fit the proxy probability function model, establish the probability distribution of the function, and obtain the expected mean and variance of each hyperparameter group; (5-4) Based on the above fitted proxy probability function, further accurately fit the proxy probability function, and gradually approach the target function by sampling in the area with large variance, so that the proxy probability function continuously approaches the target function; (5-5), repeat steps (5-3) and (5-4) until the stopping condition is reached; (5-6) Obtain the optimal hyperparameter combination through the fitted proxy probability function; (6) Substitute the data set integrated in step (3) above into the MLP neural network prediction model with the optimal hyperparameter combination for training; (7) The fatigue life value of the turbocharger pipeline is predicted by the MLP neural network prediction model trained in step (6).

2. The fatigue life prediction method of a automotive turbocharger pipeline according to claim 1, characterized in that: The specific method for obtaining the SN curve in step (1) is: (1-1) Assemble the automobile turbocharger pipeline on a tensile test bench and perform tensile tests under different displacements to obtain the number of fatigue cycle of the turbocharger pipeline under different displacements; (1-2) Calculate the stress amplitude of the turbocharger pipeline under different displacements through simulation; (1-3) The fatigue cycle number under different displacements obtained from the tensile test is fitted with the corresponding stress amplitude obtained from the simulation to obtain the SN curve, and the relationship is: , where: S represents the stress amplitude of the turbocharger pipeline; N Indicates the number of fatigue cycles of the turbocharger pipeline; m, C They are the fatigue strength index and fatigue strength coefficient of the turbocharger pipeline respectively.

3. The fatigue life prediction method of a automotive turbocharger pipeline according to claim 1, characterized in that: In the step (3), the specific method for obtaining the factors that have a great influence on the fatigue life of the turbocharger pipeline is: based on the simulation calculation model, the control variable method is used to study the influence of the structural parameters of the pipeline and the external load factors on the fatigue life of the pipeline, and the factors that have a great influence on the fatigue life of the turbocharger pipeline are analyzed.

4. The fatigue life prediction method of a automotive turbocharger pipeline according to claim 1, characterized in that: In the step (3), the condition set of the turbocharger pipeline is obtained by implementing the optimal Latin hypercube sampling method.

5. The fatigue life prediction method of a automotive turbocharger pipeline according to claim 1, characterized in that: The prediction method of step (7) is as follows: inputting the structural parameters and external loads of the new turbocharger pipeline model into the MLP neural network prediction model to calculate the fatigue life value of the turbocharger pipeline.

Images