An integrated methodology for dynamic risk prediction of thermal runaway in lithium-ion batteries
Huixing Meng,Qiaoqiao Yang,Enrico Zio,Jinduo Xing
DOI: https://doi.org/10.1016/j.psep.2023.01.021
2023-01-18
Process Safety and Environmental Protection: Transactions of the Institution of Chemical Engineers, Part B
Abstract:The risk of thermal runaway in lithium-ion battery (LIB) attracts significant attention from domains of society, industry, and academia. However, the thermal runaway prediction in the framework of system safety requires further efforts. In this paper, we propose a methodology for dynamic risk prediction by integrating fault tree (FT), dynamic Bayesian network (DBN) and support vector regression (SVR). FT graphically describes the logic of mechanism of thermal runaway. DBN allows considering multiple states and uncertain inference for providing quantitative results of the risk evolution. SVR is subsequently utilized for predicting the risk from the DBN estimation. The proposed methodology can be applied for risk early warning of LIB thermal runaway. The multi-dimensional interaction and noise interference of the original data increase the difficulty of state-of-charge (SoC) estimation. Meanwhile the existing SoC estimators prefer to estimate the SoC of consecutive time steps, which may lead to cumulative errors and data leakage. To solve these problems, a fusion network combining a multi-dimensional residual shrinkage network (MRSN) with a long short-term memory network (LSTM) is proposed for SoC estimation. Meanwhile, a sequence-to-point processing method is utilized to avoid data leakage, which is based on historical input data for estimating SoC at a single time step. Specifically, MRSN uses small sub-networks to remove redundant noise and extracts multi-scale local features on different channels to obtain stronger correlation features. LSTM uses historical inputs to obtain multi-scale temporal correlation. After that the features extracted by both networks are fused in the channel dimension to enhance the accuracy of SoC estimation. Experiments on the public dataset show that the mean absolute error (MAE) of the results is kept within 0.5 % under four different temperature conditions. Furthermore, the MAE is 0.18 % at 25 degrees, which verifies that the proposed network could significantly improve the accuracy of the estimation. Climate change and population growth require the development and operations of critical power infrastructure systems to meet the energy demand. Performance assessment of the power infrastructure usually focuses primarily on the thermodynamic analysis calculating the energy output and system efficiency. However, the energy output from the thermodynamic analysis under normal operation could not fully represent the infrastructure behavior due to the neglect of the component reliability in the energy generation system. There is a lack of consideration of the energy delivery system and its fragility with respect to external meteorological parameters (e.g., wind). In this study, the fragility of the energy-delivery system concerning wind speed is modeled and integrated to assess critical power infrastructure performance. The reliability consideration of the energy generation system is included based on the state space diagram and the frequency-balance method. The fragility of the energy-delivery system is used to derive a system reliability index for the system which is the probability that end-users will receive the energy. Besides, sensitivity analysis is conducted to investigate the influence of the important decision variables on the system performance. The study employs a small-scale power infrastructure system at the University of Connecticut as an example to demonstrate the feasibility of the performance assessment framework. The system is composed of a cogeneration plant and energy distribution systems. The results from the case study prove that the decision variables have considerable nonlinear impacts on the system performance. With the integration of fragility, the proposed approach could estimate the energy received by consumers from the generation site. With the development of fast-charging technology, fast-charging protocols have been widely used in electric vehicles. However, charging the same type of lithium-ion batteries with different fast-charging protocols has an impact on the cycle life of the battery. Considering the impact of fast-charging protocols on battery life, this paper proposes a life prediction model for lithium-ion batteries that charge with fast-charging protocols. First, the charging-based features are extracted from charge data using dilated convolutional network. Second, the discharging-based features extracted from the discharge curve are enhanced using deep neural network. Finally, the charging-based features and the discharging-based features are merged and fed into random forest regression to predict the cycle life of lithium-ion batteries. Considering that hyperparameters of the model have great influence on the prediction performance, the Bayesian optimization algorithm is used to optimize the hyperparameters of random forest regression. Two public datasets are used for -Abstract Truncated-