Predicting future capacities and remaining useful life (RUL) with uncertainty quantification is a key but challenging issue in the applications of battery health diagnosis and management. This article applies advanced machine-learning techniques to achieve effective future capacities and RUL prediction for lithium-ion (Li-ion) batteries with reliable uncertainty management. To be specific, after using the empirical mode decomposition (EMD) method, the original battery capacity data is decomposed into some intrinsic mode functions (IMFs) and a residual. Then, the long short-term memory (LSTM) submodel is applied to estimate the residual while the Gaussian process regression (GPR) submodel is utilized to fit the IMFs with the uncertainty level. Consequently, both the long-term dependence of capacity and uncertainty quantification caused by the capacity regenerations can be captured directly and simultaneously. Experimental aging data from different batteries are deployed to evaluate the performance of proposed LSTM+GPR model in comparison with the solo GPR, solo LSTM, GPR+EMD, and LSTM+EMD models. Illustrative results demonstrate the combined LSTM+GPR model outperforms other counterparts and is capable of achieving accurate results for both 1-step and multistep ahead capacity predictions. Even predicting the RUL at the early battery cycle stage, the proposed data-driven approach still presents good adaptability and reliable uncertainty quantification for battery health diagnosis.

Battery calendar aging prediction is of extreme importance for developing durable electric vehicles. This paper derives machine learning-enabled calendar aging prediction for lithium-ion batteries. Specifically, the Gaussian process regression (GPR) technique is employed to capture the underlying mapping among capacity, storage temperature, and SOC. By modifying the isotropic kernel function with an automatic relevance determination (ARD) structure, high relevant input features can be effectively extracted to improve prediction accuracy and robustness. Experimental battery calendar aging data from nine storage cases are utilized for model training, validation, and comparison, which is more meaningful and practical than using the data from a single condition. Illustrative results demonstrate that the proposed GPR model with ARD Matern32 (M32) kernel outperforms other counterparts and can achieve reliable prediction results for all storage cases. Even for the partial-data training test, multi-step prediction test and accelerated aging training test, the proposed ARDbased GPR model is still capable of excavating the useful features, therefore offering good generalization ability and accurate prediction results for calendar aging under various storage conditions. This is the first known data-driven application that utilizes the GPR with ARD kernel to perform battery calendar aging prognosis.

Lithium-ion battery manufacturing is a highly complicated process with strongly coupled feature interdependencies; a feasible solution that can analyze feature variables within manufacturing chain and achieve reliable classification is, thus, urgently needed. This article proposes a random forest (RF)-based classification framework, through using the out of bag predictions, Gini changes, as well as predictive measure of association (PMOA), for effectively quantifying the importance and correlations of battery manufacturing features and their effects on the classification of electrode properties. Battery manufacturing data containing three intermediate product features from the mixing stage and one product parameter from the coating stage are analyzed by the designed RF framework to investigate their effects on both the battery electrode active material mass load and porosity. Illustrative results demonstrate that the proposed RF framework not only achieves the reliable classification of electrode properties, but also leads to the effective quantification of both manufacturing feature importance (FI) and correlations. This is the first time to design a systematic RF framework for simultaneously quantifying battery production FI and correlations by three various quantitative indicators, including the unbiased FI, gain improvement FI, and PMOA, paving a promising solution to reduce model dimension and conduct efficient sensitivity analysis of battery manufacturing.

Accurate prediction of batteries’ future degradation is a key solution to relief users’ anxiety on battery lifespan and electric vehicle’s driving range. Technical challenges arise from the highly nonlinear dynamics of battery aging. In this paper, a feed-forward migration neural network is proposed to predict the batteries’ aging trajectories. Specifically, a base model that describes the capacity decay over time is first established from the existed battery aging dataset. This base model is then transformed by an input-output slope-and-bias-correction (SBC) method structure to capture the degradation of target cell. To enhance the model’s nonlinear transfer capability, the SBC-model is further integrated into a four-layer neural network, and easily trained via the gradient correlation algorithm. The proposed migration neural network is experimentally verified with four different commercial batteries. The predicted RMSEs are all lower than 2.5% when using only the first 30% of aging trajectories for neural network training. In addition, illustrative results demonstrate that a small size feed-forward neural network (down to 1-5-5-1) is sufficient for battery aging trajectory prediction.

Lithium-ion (Li-ion) batteries undergo complex electrochemical and mechanical degradation. This complexity is pronounced in applications such as electric vehicles, where highly demanding cycles of operation and varying environmental conditions lead to non-trivial interactions of ageing stress factors. This work presents the framework for an ageing diagnostic tool based on identifying and then tracking the evolution of model parameters of a fundamental electrochemistry-based battery model from non-invasive voltage/current cycling tests. In addition to understanding the underlying mechanisms for degradation, the optimisation algorithm developed in this work allows for rapid parametrisation of the pseudo-two dimensional (P2D), Doyle-Fuller-Newman, battery model. This is achieved through exploiting the embedded symbolic manipulation capabilities and global optimisation methods within MapleSim. Results are presented that highlight the significant reductions in the computational resources required for solving systems of coupled non-linear partial differential equations.