Detecting thermal faults is critical to the safety of lithium-ion batteries. This paper, therefore, proposes a neural network-based approach. The approach relies on the long short-term memory (LSTM) neural network, in conjunction with an alteration to the walk-forward technique, to accurately estimate the surface temperature of the cell. It also relies on a residual monitor to detect the faults in real-time. This data-driven method is introduced to expand the available options in thermal fault detection. It offers an easy-to-implement option that does not require expert understanding in battery physics, complex mathematical modeling, and tedious parameter tuning processes. The experimental results demonstrate that this approach can detect thermal faults accurately. It is adaptive to different battery chemistries and form factors, and thanks to its online training capability, it can also automatically retrain itself to capture changes in the battery over time.

The physics-informed neural network (PINN) has drawn much attention as it can reduce training data size and eliminate the need for physics equation identification. This paper presents the implementation of a PINN with adaptive normalization in the loss function to predict lithium-ion battery cell temperature. In particular, the PINN was trained with the actual battery test data, and a lumped capacitance lithium-ion battery thermal relationship was applied to the loss function with the addition of a pre-layer and connection layer to the neural network architecture. The PINN architecture shows the most accurate battery temperature prediction compared with the fully connected neural network (FCN) and its variants evaluated in this study. The proposed PINN architecture has a mean square prediction error of 0.05 °C with a limited number of training data and without battery thermal model identification.

All-solid-state batteries (ASSBs) are considered to be the next generation of lithium-ion batteries. Physics-based models (PBMs) can effectively simulate the internal electrochemical reactions and provide critical internal states for battery management. In order to promote the onboard applications of PBMs for ASSBs, in this article, the parameter sensitivity of a typical PBM is analyzed, and a joint estimation method for states and parameters based on sigma-point Kalman filtering (SPKF) is proposed. First, to obtain accurate sensitivity analysis results, approaches from different principles, including local sensitivity, elementary effect test, and variance-based methods, are applied. Then, for the battery model based on partial differential equations, a nonlinear state-space model is constructed by using the finite-difference discretization method. Finally, the SPKF algorithm is employed to conduct the joint estimation of model parameters and lithium-ion concentrations. The results from constant current and dynamic cycles show that two parameters, namely maximum lithium-ion concentration and minimum lithium-ion concentration, have the most influence on the model results. The joint estimation method is validated in three different cases, and the mean absolute errors of the estimated voltage and state of charge (SOC) are below 2.1 mV and 1.5%, respectively.

This paper proposes a predictive equivalent consumption minimization strategy (P-ECMS) for a plug-in hybrid electric vehicle (PHEV), assuming the availability of two levels of traffic information. The two levels of traffic information include 1) segmented traffic information available in mobile mapping applications, and 2) detailed velocity information, possibly obtained by short-term speed forecasting. Battery state-of-charge (SOC) reference waypoints are obtained by a simplified speed profile constructed from the segmented traffic information. The proposed P-ECMS adjusts its co-state or equivalence factor based on the difference between the future SOC obtained from short-horizon prediction and a future reference SOC. The benefits of the proposed P-ECMS are evaluated through vehicle simulations on a specified trip, compared against an adaptive ECMS (A-ECMS) with the same control parameters minus prediction. Simulation results on the trip show that the proposed P-ECMS can achieve a 9.7% reduction in the corresponding fuel consumption on average, as compared to the A-ECMS across different combinations of initial co-state and control gain. Moreover, its performance is robust to variations in the initial co-state, the co-state update rate, and the prediction horizon, as well as inaccuracies in the velocity prediction. Compared to the A-ECMS, the standard deviation in fuel consumption with the P-ECMS can be reduced by 96% from 84.1 to 3.4 grams.