| According to the data of the National Development and Reform Commission,the domestic ownership of new energy vehicles had reached 10.1 million by the end of June 2022,accounting for3.23% of the total number of vehicles,among which,the ownership of pure electric vehicles reached 8.104 million,accounting for 80.93% of the total proportion.However,under these figures showing the huge market potential of electric vehicles,the battery system of electric vehicles also faces great challenges,specifically,the poor detection and diagnosis of power battery pack faults,the lack of detection methods for various types of minor battery faults,and the misdiagnosis between different types of faults.These problems have led to a number of battery combustion accidents in vehicles,which seriously threaten people’s lives and property safety.Therefore,this paper conducts a study on the following.Firstly,the working principle of lithium-ion battery is used as the starting point to start the research,and the evolution mechanism of common failures of lithium-ion battery is studied.We design and implement a fault experiment plan,build a battery pack experimental platform for fault detection and diagnosis,simulate common faults of lithium-ion batteries,and lay the foundation for the subsequent research on fault detection and diagnosis methods based on deep learning.Then for the problem of lithium-ion battery pack fault detection,this chapter proposes a correlation coefficient combined with neural network for lithium-ion battery fault diagnosis scheme by using the advantages of Spearman’s rank correlation coefficient.The method uses Spearman’s rank correlation coefficient to dimensionless and normalize the battery fault features,providing a new solution to the problem of generalization ability when the amount of fault data is insufficient for traditional models.Then,an improved self-coding neural network is used to reconstruct the battery voltage sequence and detect its abnormality degree to determine whether there is a fault in the battery pack,and to determine the time of fault occurrence according to the range involved in the abnormality,based on which a dynamically adjusted threshold determination method is designed to achieve accurate fault detection and localization.And through simulation and actual battery pack experimental platform for testing,it is verified that the detection scheme can achieve accurate and rapid detection of all kinds of minor faults in battery packs,which is more effective than the detection by traditional methods.Finally,the lithium-ion battery pack fault detection model is only able to provide early warning of the occurrence of a fault in the battery pack,but cannot diagnose what kind of fault has occurred.In this chapter,based on the fault cell locations detected in the previous chapter and intercepting the corresponding fault data fragments,a battery pack fault diagnosis algorithm based on CRDAN neural network is proposed by improving the characteristics of domain adaptation neural network combined with attention mechanism to obtain good cross-domain diagnosis performance by using fault enhancement mechanism to target the relative importance of generic battery pack fault features.To solve the problem of how to effectively extract the fault features of multiple cells of a battery pack,this chapter designs a feature extraction strategy with channel and time attention mechanisms,which can effectively use the channel and time attention mechanisms to guide the extraction of fault features.Finally,considering the influence of different operating environments on the performance degradation of Li-ion batteries,diagnostic tests are conducted under various operating environments of FUDS,UDDS,and US06.After the experimental analysis,the proposed diagnosis method is validated on the self-built battery fault data set and the actual joint detection and diagnosis in this paper,and the results show that the accuracy and effectiveness of the method in this paper achieve the expected results. |
PDF Full Text Request