Research On Early Warning Methods And Technologies For Thermal Runaway Of Lithium-ion Battery

In order to fulfill the requirements for extremely early warning of lithium-ion battery on thermal runaway germinal and occurrence stages.Based on a combination method of theories and experiments,a dynamic warning system for thermal runaway of lithium-ion battery is established,the thermal runaway warning model linked countdown index and thermal runaway level is constructed,and the generalisation ability of the model is also verified on fresh samples of thermal runaway in the same-material system.(1)It is suggested to use high-capacity ternary lithium-ion power battery thermal runaway eigenvalues to build an early warning system.A dynamic early warning system with the integrator of the battery multi-point temperature,hydrogen,carbon monoxide,and hydrocarbon gases is built with the combination of heat generation,gas production mechanism,and experimental characteristic parameters.The time series of the peak of each key indicator is shown through the thermally induced runaway experiments under three states of charge(SOC)of 100%,60%,and 30%.A dynamic early warning system with the integration of heat and gas production is established and the reliability of the indicators is verified;then the classification logic of the warning level and the input and output characteristics of the warning model are analyzed in depth,the redundant information of the early warning system is eliminated,the temporal and spatial characteristic index system composed of heat and gas production is optimized,and the total data set of the thermal runaway early warning model integrating training,testing and verification is created.(2)To meet the criteria for the proposed extremely early warning,a model and rating system for lithium-ion battery thermal runaway are constructed.Based on the optimization of the number of neurons with activation function of the hidden layer of the optimal artificial neural network(ANN)in this thesis,the iteration accuracy of the thermal runaway characteristics data set is enhanced;with the ANN regression warning model trained by Levenberg-Marquardt(LM)algorithm,the thermal runaway warning model based on countdown index is created,the correlation between thermal runaway heat and gas production and the countdown is analysed and the warning time lead time is accurately assessed based on the thermal runaway countdown model;fusing the ANN algorithm and extreme gradient boosting tree(XGBoost)algorithm,the classification early warning level is greatly optimized.Finally,the indicators of Weight and Cover are used to assess the significance of the index characteristic parameters,and a high-precision thermal runaway early warning technology model is developed by integrating the countdown time series and indicator threshold proportions.(3)A method to validate the generalization capability of the early warning model based on thermal runaway data under the same battery material system is investigated.In this thesis,the experimental data of NCM523 square lithium-ion batteries thermal runaway at 100%,60% and 30% SOC under atmospheric pressure environment are used to extract the same heat and gas production data as the experimental analysis of the early warning indicator system,to construct a fresh sample validation dataset,and to analyse and validate the generalisation capability of the countdown warning model and the hierarchical warning model on this dataset.In this thesis,the method and technique for lithium-ion battery thermal runaway early warning are investigated.The risk of thermal runaway on the germinal stage is effectively identified,and as a result,a four-level early warning mechanism is established,which serves as a reference for the research on lithium-ion battery thermal runaway early warning.