Abstract:Accurate lifetime prediction of lithium-ion batteries in the early cycles is critical for timely failure warning and effective quality grading. Convolutional neural network (CNN), with excellent performance in feature extraction, has gained increasingly attentions in battery prognostics. However, since degradation test normally takes years to complete, employing end-to-end CNNs directly for battery lifetime prediction is impractical due to the limited number of available training samples and the scarcity of features in the early cycles. Instead of directly feeding the raw data, in this work, we propose to use graphical features for early lifetime prediction. Three feature curves, including capacity-voltage curve, incremental capacity curve, and capacity difference curve are used to construct graphical features. Specifically, the incremental capacity curve and capacity difference curve are derived from capacity-voltage curve, aiming to extract more information from both intra-cycle and inter-cycle perspectives. The evolution patterns of these feature curves over the initial 100 cycles show evident correlations with battery lifetime, and are termed as the graphical features. The three graphical features, after some proper transformation, are stacked into a three-channel image before feeding to the CNN model. Five classical CNNs, with different structures and key parameters, are investigated for battery lifetime prediction. Comparative experiments are conducted to study the influence of different feature combinations, voltage segments, and discharge cycles on the prediction performance. Experimental results demonstrate that simple CNNs with only a few convolutional layers can achieve satisfying prediction results. Additionally, networks with rectified linear unit are shown to outperform those with other activation functions.

What problem does this paper attempt to address?

The paper mainly discusses the problem of early-life prediction of lithium-ion batteries. In battery management systems, accurate early-life prediction of batteries is crucial as it can provide early warning of failure and effectively grade battery quality. However, directly using end-to-end convolutional neural networks (CNNs) for life prediction is not practical due to the limited number of available training samples and the scarcity of features in early cycles, as battery degradation testing typically takes several years to complete. The study proposes a new method for predicting early-life of batteries using graphical features. Specifically, three types of feature curves, namely capacity-voltage curve, incremental capacity curve, and capacity difference curve, are constructed from the perspective of internal and external cycles to extract battery degradation information. These feature curves exhibit a clear correlation with battery life in the initial 100 cycles. After appropriate transformation, these three graphical features are combined into a three-channel image input to the CNN model. The paper investigates five classical CNN structures and analyzes the impact of different feature combinations, voltage segments, and discharge cycles on prediction performance. Experimental results show that even with only a few convolutional layers, a simple CNN model can achieve satisfactory prediction results, with networks using the ReLU activation function performing better than other activation functions. The paper also compares the application of model-based methods and machine learning methods in the prediction of remaining useful life (RUL) of batteries, pointing out that while model-based methods can achieve good results, they are complex and require in-depth understanding of battery internal mechanisms. In contrast, machine learning methods can directly extract features from battery data for prediction. Although previous research has achieved success in predicting the late-life of batteries, methods relying solely on capacity data often perform poorly in early stages. Therefore, the proposed method in the paper aims to improve the accuracy of early-life prediction by utilizing more information such as voltage, current, and temperature. The paper utilizes a dataset of 124 lithium iron phosphate battery samples provided by MIT and performs early-life prediction using the CNN model based on graphical features. The study finds that most samples have battery life durations between 450 and 1050 cycles, while early-stage and late-stage samples each account for approximately 15%. By conducting comparative experiments on different features, voltage segments, and discharge cycles, the paper evaluates the prediction performance of different CNN models. In conclusion, the goal of this paper is to address how to effectively predict the early-life of lithium-ion batteries using early battery data and graphical features through convolutional neural networks, in order to improve the accuracy and timeliness of predictions.

Early prediction of battery lifetime based on graphical features and convolutional neural networks

An Interpretable Semi-Supervised Learning Approach for Battery Lifespan Early Prognostic

Early prediction of lithium-ion battery lifetime via a hybrid deep learning model

Fast battery capacity estimation using convolutional neural networks

A machine-learning prediction method of lithium-ion battery life based on charge process for different applications

Early-stage lifetime prediction for lithium-ion batteries: A deep learning framework jointly considering machine-learned and handcrafted data features

Prediction of the Remaining Useful Life of Lithium-Ion Batteries Based on the 1D CNN-BLSTM Neural Network

A deep learning approach incorporating attention mechanism and transfer learning for lithium-ion battery lifespan prediction

Lithium-ion battery degradation trajectory early prediction with synthetic dataset and deep learning

Rapid Ultracapacitor Life Prediction with a Convolutional Neural Network

A Hybrid Machine Learning Model for Battery Cycle Life Prediction with Early Cycle Data

A two-stage integrated method for early prediction of remaining useful life of lithium-ion batteries

Deep neural network battery life and voltage prediction by using data of one cycle only

Deep Learning Powered Lifetime Prediction for Lithium-Ion Batteries Based on Small Amounts of Charging Cycles

Adaptable capacity estimation of lithium-ion battery based on short-duration random constant-current charging voltages and convolutional neural networks

Remaining Useful Life Prediction of Lithium Battery Based on Sequential CNN–LSTM Method

Early prediction of battery lifetime via a machine learning based framework

A two-stage deep learning framework for early-stage lifetime prediction for lithium-ion batteries with consideration of features from multiple cycles

Early prediction of cycle life for lithium-ion batteries based on evolutionary computation and machine learning

A Hybrid Convolutional Neural Network-Long Short Term Memory for Discharge Capacity Estimation of Lithium-Ion Batteries