CN-122020313-A - Construction method of trusted basic working condition library of electric drive system

CN122020313ACN 122020313 ACN122020313 ACN 122020313ACN-122020313-A

Abstract

The invention discloses a method for constructing a reliable basic working condition library of an electric drive system, which comprises the steps of obtaining original operation data of the electric drive system, preprocessing the original operation data, dividing operation fragments, constructing a damage intensity characteristic matrix based on the operation fragments according to a failure physical model of a core component of the electric drive system, carrying out dimension reduction processing on the damage intensity characteristic matrix to obtain low-dimension characteristic data, carrying out cluster analysis by adopting a Gaussian mixture model and combining an optimal cluster number judgment strategy to obtain an operation condition classification result, determining the minimum sampling mileage of the basic working condition library by utilizing a variable sliding time window method, and constructing the basic working condition library by adopting a classification random sampling and consistency inspection method based on a working condition proportion constraint according to the minimum sampling mileage. The invention can obviously reduce the data acquisition cost and shorten the reliability test period and improve the accuracy and efficiency of the reliability evaluation of the electric drive system on the premise of ensuring the comprehensiveness and the randomness of the coverage of the working conditions.

Inventors

ZHANG DONGDONG
ZHAO LIHUI
Li Lingzhou
YAO TAO
WANG ZHEN
KONG ZHIGUO
LONG XU
Xia Huaixiao

Assignees

上海理工大学

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The method for constructing the trusted base working condition library of the electric drive system is characterized by comprising the following steps of: acquiring original operation data of an electric drive system, preprocessing the original operation data and dividing operation fragments; Constructing a damage intensity feature matrix based on the operation segment according to a failure physical model of the electric drive system core component; performing dimension reduction treatment on the damage intensity feature matrix to obtain low-dimension feature data; According to the low-dimensional characteristic data, carrying out cluster analysis by adopting a Gaussian mixture model and combining an optimal cluster number judgment strategy to obtain an operation condition classification result; Determining the minimum sampling mileage of a basic working condition library by utilizing a variable sliding time window method according to the operation condition classification result; and constructing a basic working condition library by adopting a classified random sampling and consistency checking method based on the working condition proportion constraint according to the minimum sampling mileage.
2. The method of claim 1, wherein the step of determining the position of the substrate comprises, The process of preprocessing the original operation data and dividing the operation fragments comprises the following steps: acquiring original operation data, and carrying out resampling, linear interpolation and outlier rejection processing on the original operation data; defining a continuous running process from an initial zero value to a next zero value of the vehicle speed as a segment according to the preprocessed data; Dividing the segment into an operation segment or an idle segment according to the maximum vehicle speed in the segment, and extracting the operation segment as the operation segment.
3. The method of claim 1, wherein the step of determining the position of the substrate comprises, The process for constructing the damage intensity characteristic matrix based on the operation fragments comprises the following steps: According to failure physical models of shaft parts, gear parts, bearing parts, stator windings and IGBT devices of the electric drive system, respectively calculating pseudo damage of each part under the operation segment; Calculating the damage intensity of each part according to the pseudo damage and the mileage of the corresponding operation section; and constructing and obtaining the damage intensity characteristic matrix according to the damage intensity of each part.
4. The method of claim 1, wherein the step of determining the position of the substrate comprises, The damage intensity feature matrix is subjected to dimension reduction treatment, and the process of obtaining low-dimension feature data comprises the following steps: Constructing a weighted k-adjacency graph according to the damage intensity characteristics and calculating the connection weight between high-dimensional data points; And optimizing the low-dimensional coordinate representation by minimizing cross entropy according to the connection weight, and mapping the high-dimensional damage intensity characteristic to a low-dimensional space by adopting a uniform manifold approximation and projection algorithm to obtain the low-dimensional characteristic data.
5. The method of claim 1, wherein the step of determining the position of the substrate comprises, The process for obtaining the operation condition classification result comprises the following steps: According to the low-dimensional characteristic data, calculating the davison baudines indexes under different clustering numbers; determining the optimal cluster number according to the minimum value of the davison bauding index; calculating posterior probability of each sample by using a Bayes formula according to the optimal cluster number; and clustering by adopting a Gaussian mixture model according to the posterior probability, and iteratively optimizing model parameters by an expected maximization algorithm, and dividing the operation fragments into corresponding Gaussian components to obtain the operation condition classification result.
6. The method of claim 1, wherein the step of determining the position of the substrate comprises, The process of determining the minimum sampling mileage of the base working condition library by utilizing the variable sliding time window method comprises the following steps: Dividing the multidimensional load data into sample sets with different window sizes by adopting a variable sliding time window method; According to the sample set, calculating variation coefficients in the time domain, the frequency domain and the damage domain dimension respectively; Constructing a relative recurrence fitting model according to the variation coefficient; And calculating the minimum sampling mileage of different users according to the relative recurrence fitting model, fitting the cumulative probability distribution model, and adopting the user level of 50% as the minimum sampling mileage of the basic working condition library.
7. The method of claim 1, wherein the step of determining the position of the substrate comprises, The process for constructing the basic working condition library by adopting the classified random sampling and consistency checking method based on the working condition proportion constraint comprises the following steps: calculating the number of sampling fragments corresponding to various working conditions according to the proportion relation between the minimum sampling mileage and the total mileage of the overall sample; and according to the number of the sampling fragments, independent random sampling is implemented in each category by adopting a classification random sampling and consistency checking method, and sample working condition fragment combinations meeting constraint conditions are extracted to construct the basic working condition library.
8. The method of claim 1, wherein the step of determining the position of the substrate comprises, The construction of the basic working condition library further comprises the following steps: According to the basic working condition library and the overall working condition library, performing double-sample Kolmogorov-Smirnov test, Z test and F test to obtain test results; And calculating consistency statistics of the basic working condition library and the overall working condition library on time domain features, frequency domain features and damage domain features according to the test result.
9. The method of claim 8, wherein the step of determining the position of the first electrode is performed, The two-sample Kolmogorov-Smirnov test, Z test and F test were followed by: calculating joint probability distribution errors according to the rotation speed-torque joint probability distribution of the basic working condition library and the overall working condition library; And constructing an accumulated probability distribution curve according to the unit damage intensity of each key component in the basic working condition library and the overall working condition library, and comparing and analyzing the difference degree.
10. The method of claim 1, wherein the step of determining the position of the substrate comprises, The failure physical model includes: The method comprises a shaft part fatigue damage model, a gear contact stress and bending stress calculation model, a bearing service life calculation model, a stator winding insulation aging model based on an Arrhenius equation and an IGBT thermal cycle fatigue model based on a Coffin-Manson model, which are corrected based on a Goodman straight line criterion.

Description

Construction method of trusted basic working condition library of electric drive system Technical Field The invention belongs to the field of automobile electric drive systems, and particularly relates to a method for constructing a trusted basic working condition library of an electric drive system. Background The electric drive system is used as a core power system of the new energy automobile, and the reliability of the electric drive system directly influences the performance and safety of the whole automobile. Currently, the construction of the reliability test working condition of an electric drive system mainly depends on actual measurement data acquisition of a full life cycle user road load spectrum, and the driving data are obtained by road tests in multiple scenes such as cities, suburbs, highways, mountain roads and the like, and a proportional working condition library is constructed based on statistical distribution. Meanwhile, with the development of artificial intelligence technology, an unsupervised clustering algorithm (such as hierarchical clustering, spectral clustering, t-SNE clustering, K-means and the like) is widely applied to working condition extraction and analysis, and working condition classification is realized through processing kinematic characteristic parameters such as vehicle speed, acceleration, rotating speed, torque and the like, so that a data base is provided for a reliability test. However, the existing working condition library constructed based on proportionality statistics lacks the capability of simulating random alternation and uncertainty of working conditions in a real driving environment, is difficult to cover a small probability scene (such as sudden rapid acceleration, long-time idling and the like), and lacks sufficient verification of the relevance between typical road working conditions simulated by a test field and actual user driving behaviors, environments and road surface working conditions, and in addition, large-scale road data acquisition needs to input a large amount of manpower, material resources and time cost, so that the test efficiency is low. On the other hand, the characteristic parameters constructed by the existing clustering method mainly pay attention to kinematic indexes such as speed, acceleration, torque fluctuation and the like, the inherent association with failure mechanisms of key components (such as shafts, gears, bearings, windings, IGBT and the like) of an electric drive system cannot be fully considered, the damage characteristics of all the components are difficult to accurately represent, and the constructed working condition library is insufficient in reliability evaluation. Disclosure of Invention In order to solve the problems that the reliability test working condition of the current electric drive system depends on large-scale data acquisition, the redundancy of real vehicle operation data is high and the coverage of the operation working condition is incomplete, the invention provides a method for constructing a trusted basic working condition library of the electric drive system, which comprises the following steps: acquiring original operation data of an electric drive system, preprocessing the original operation data and dividing operation fragments; Constructing a damage intensity feature matrix based on the operation segment according to a failure physical model of the electric drive system core component; performing dimension reduction treatment on the damage intensity feature matrix to obtain low-dimension feature data; According to the low-dimensional characteristic data, carrying out cluster analysis by adopting a Gaussian mixture model and combining an optimal cluster number judgment strategy to obtain an operation condition classification result; Determining the minimum sampling mileage of a basic working condition library by utilizing a variable sliding time window method according to the operation condition classification result; and constructing and obtaining a basic working condition library by adopting a classified random sampling and consistency checking method based on the working condition proportion constraint according to the minimum sampling mileage. Preferably, the process of preprocessing the original operation data and dividing the operation fragments includes: acquiring original operation data, and carrying out resampling, linear interpolation and outlier rejection processing on the original operation data; defining a continuous running process from an initial zero value to a next zero value of the vehicle speed as a segment according to the preprocessed data; Dividing the segment into an operation segment or an idle segment according to the maximum vehicle speed in the segment, and extracting the operation segment as the operation segment. Preferably, the process of constructing the damage intensity feature matrix based on the running segment includes: According to failure physical models o