Research On Power Battery Multi-Fault Diagnosis Strategy Based On Data-Driven

Due to the excessive consumption of fossil fuels and the resulting serious environmental pollution,new energy technologies are gaining more and more attention.Countries are promoting the development of electric vehicles and implementing relevant policies,especially China.Meanwhile,lithium-ion batteries have become one of the main power sources of electric vehicles due to their advantages of high energy density,fast charging speed,and high discharge current,and their growth has been explosive.However,the safety issues of power batteries have gradually emerged.In recent years,electric vehicles have experienced safety accidents,such as spontaneous combustion and explosions during charging,when stationary,or while driving.The main cause of these safety accidents is thermal runaway in electric vehicles,which results from faults that are not detected on time.Therefore,fault diagnosis technology is a key technology for power batteries and a guarantee for the safe operation of electric vehicles.The ability to diagnose faults and hidden dangers on time has become a core concern for battery researchers and vehicle manufacturers.Thus,this article aims to conduct research on the fault diagnosis of power batteries,with the following main research objectives:(1)This article analyzes the common types of battery faults,their causes,and their evolution process,and summarizes two aspects that need to be improved in the field of fault diagnosis.The first aspect is to improve the prediction of battery faults.Most of the existing researches about fault diagnosis are based on diagnosing the already occurred faults.However,safety can only be greatly improved by predicting faults in advance.The second aspect is to improve multi-level and multi-fault diagnosis.Because every level in the battery system may have multiple types of faults.Currently,most researches focus on the Battery Cell Level(BCL),with fewer studies on the Battery Module Level(BML),and even fewer on the Battery Pack Level(BPL).(2)To address the need to improve fault prediction in the field of battery fault diagnosis,a fault prediction method based on a Convolutional Neural Network-Long Short-Term Memory(CNN-LSTM)network combined with correlation coefficient is proposed in this paper.Data support is required to verify the method,and lithium-ion battery(INP2714891A)is used as the research object.The necessary test platform and fault scheme are designed in the laboratory.First,historical voltage data is used as input to the CNN-LSTM voltage prediction model for training based on the fault data,and the optimal prediction model is obtained.The feature extraction ability of the CNN is utilized to improve the accuracy of the LSTM in processing time series prediction.Then,the saved model is used to predict the voltage of all battery cells,and the predicted values of each cell are combined with their partial historical data to form corresponding new sequences.Finally,the correlation coefficient is used to calculate the correlation between new sequences of adjacent batteries,and the results are analyzed to diagnose faults.(3)To address the need to improve multi-level and multi-fault diagnosis in the field of battery fault diagnosis,a multi-fault diagnosis method is proposed in this paper for battery system based on a Non-Redundant Interleaved Measurement Circuit(NRIMC)combined with Improved Fuzzy Entropy(IFuzzy En).The method is based on the experimental data of lithium-ion battery(NCR18650PF),and at the same time,in order to avoid the risk of partial experiments and reduce the experimental cost,MATLAB/Simulink is used to simulate the Equivalent Circuit Model(ECM)of the battery cell and build the battery system model of BPL,followed by fault simulation experiments.The voltage sensors in the battery system are connected using the NRIMC rules to obtain voltage data.Then,the IFuzzy En is used to calculate the abnormality of the voltage sequence.Finally,the multi-fault diagnosis method is used to locate and differentiate the types and degree of the faults in the battery system accurately and quickly.The functions of this method include fault location,fault type differentiation,and fault degree judgment.In fault analysis,the diagnosis of BPL faults is performed first.Only when a fault occurs in the BPL,the diagnosis of BML faults is conducted.Similarly,only when a fault occurs in the battery module,the diagnosis of BCL faults is performed.The number of abnormal voltage sequences at the same time is used to differentiate the fault types,and the size of the abnormal values is used to determine the fault degree.In summary,as the main research work of this study,two fault diagnosis methods are proposed and various experiments are conducted to verify their accuracy and reliability.This study has profound significance for exploring battery fault diagnosis technology.If the proposed methods can be applied to actual battery management systems,it will greatly improve the safety performance of electric vehicles.