Aging States Estimation and Monitoring Strategies of Li-Ion Batteries Using Incremental Capacity Analysis and Gaussian Process Regression

Moritz Landwehr, Patrick Hoher and Johannes Reuter · 2026

Read Aging States Estimation and Monitoring Strategies of Li-Ion Batteries Using Incremental Capacity Analysis and Gaussian Process Regression by Moritz Landwehr, Patrick Hoher and Johannes Reuter (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-26155.txt.

Introduction

Landwehr, Hoher, and Reuter estimate battery aging from as little as one diagnostic cycle. Many real uses give only sparse data on a cell. Their method reads features from incremental capacity curves. A multi-model Gaussian process ensemble turns those features into health estimates. It also reports uncertainty. On the Oxford aging dataset, state of health error is just 1.3 percent. Remaining life is harder, with 5.3 percent error. Their monitoring plan needs about four checks over a cell's life. That life spans 3,300 to 5,000 cycles.

Catalog record

CREATORMoritz Landwehr, Patrick Hoher and Johannes Reuter
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38425
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.26155

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Aging States Estimation and Monitoring Strategies of Li-Ion Batteries Using Incremental Capacity Analysis and Gaussian Process Regression Moritz Landwehr, Patrick Hoher and Johannes Reuter Abstract Existing approaches for battery health forecasting often rely on extensive cycling histories and continuously monitored cells. In contrast, many real-world scenarios provide only sparse information, e.g. a single diagnostic cycle. In our study, we investigate state of health (SoH) - and remaining useful life (RUL) estimation of previously unseen lithium-ion cells, relying on cycling data from begin of life (BOL) to end of life (EOL) of multiple similar cells by using the publicly available Oxford battery aging dataset. The estimator applies incremental capacity analysis (ICA) -based feature extraction in combination with data-efficient regression methods. Particular emphasis is placed on a multi-model Gaussian process regression ensemble approach (GPR n ) , which also provides uncertainty quantification. Due to a rather cell invariant behaviour, the mapping of ICA features to SoH estimation is highly precise and points out a normalized mean absolute error (NMAE) of 1.3%. The more cell variant mapping to RUL estimation is challenging, reflecting in a NMAE of 5.3%. Using the estimation results, a RUL monitoring strategy is derived. The objective is to safely operate a battery cell from BOL to EOL by only taking sparse diagnostic measurements. On average, only four diagnostic measurements are required during a cell’s lifetime of 3300 to 5000 cycles. \phmLicenseFootnote Moritz Landwehr 1 Introduction 1.1 Motivation Lithium-ion batteries are a key technology for electric vehicles, stationary storage, and portable electronics and their aging directly impacts system reliability, safety, and cost. Capacity fade and resistance increase reduce usable energy and power, while uncontrolled degradation can lead to unplanned downtime and safety concerns. Accurate estimation of SoH and RUL is therefore essential for predictive maintenance, lifetime extension, and economic decision-making, especially in fleet and second-life applications. In many practical scenarios, however, only a few diagnostic cycles are available instead of continuous, long-term monitoring of individual cells. Methods that rely on dense cycling histories are thus difficult to apply, which creates a need for data-efficient approaches that extract information-rich features from sparse measurements and provide reliable SoH - and RUL estimates with quantified uncertainty. 1.2 Battery Modelling and Estimation Methods In demirci_review_2024 and schmitt_state_2022 an overview about different methods of SoH estimation is shown, whereas RUL estimation methods are grouped and summarized in elmahallawy_comprehensive_2022 . Both, SoH - and RUL estimation, deal with long term dynamic behaviour of battery cells, which is in contrast to state of charge (SoC) estimation. The methods for SoH - and RUL estimation can be grouped in model-based and experimental methods. Regarding model-based methods, battery aging can be modelled using a broad spectrum of approaches with different complexity, accuracy and computational intensity hamar_data-driven_2024 . Most widespread modelling approaches are electro-chemical models santhanagopalan_review_2006 , electric equivalent circuit models guo_review_2021 , electrochemical impedance models jossen_fundamentals_2006 , open circuit potential models, and data-driven models. For SoH estimation there are also several experimental methods, dividing in direct measurement methods, like impedance measurement and coulomb counting, and indirect analysis methods, like ICA and differential voltage analysis (DVA) ( demirci_review_2024 ) . Electro-chemical models typically require extensive parametrization, detailed knowledge of internal cell chemistry, and significant computational effort, which limits their scalability and applicability across different cell types and operating conditions. For electric equivalent circuit models, the aging behaviour of the circuit elements has to be mathematically described and optimized using a large, high resolution dataset. Data-driven, specifically machine learning (ML) -approaches, have therefore gained increasing attention dar_comprehensive_2025 . 1.3 Machine Learning in Battery State Estimation al-hashimi_decade_2025 review, that different neural networks topologies, like fully connected feedforward neural networks (FFNNs) , long short-term neural networks (LSTMs) , support vector regressions (SVRs) , Gaussian process regressions (GPRs) , and deep learning architectures have been successfully applied to SoH - and RUL estimation. Mainly voltage, current, temperature, and capacity over time from charging and discharging cycles are used as regressor inputs. They are comparably easy to measure. Intuitively, these features can be directly fed into one of the previously mentioned regressors. However it is a challenge to map the sparse information density of raw measurement data to meaningful features, describing the battery aging process. In the literature there are some examples given. In jha_itransformer_2024 a deep learning transformer network directly estimates the RUL with time based features as inputs. Also nguyen_van_estimation_2023 processes time based input data with a LSTM for SoH estimation. There are also regressors, like convolutional neural networks (CNNs) , including inherently feature extraction zhou_battery_2025 . All this approaches are based on observing batteries from BOL to a certain age as training data source, to afterwards predict the development of the aging until EOL . In contrast, we deal with the task of a SoH - and RUL estimation of previously unseen cells using cycling data from BOL to EOL of multiple similar cells. In this case, due to cell-to-cell variance the direct processing of sparse information dense time based features is critical. Here ICA has emerged as a powerful non-destructive diagnostic technique to extract high information dense aging-related features. 1.4 Battery State Estimation Using ICA As she_battery_2023 describe, the ICA method transforms voltage plateaus and inflection points on the voltage-capacity curves to identifiable peaks and valleys on the corresponding incremental capacity (IC) curves. The characteristic shape of the IC curves directly corresponds to degradation mechanisms, like loss of lithium inventory (LLI) , loss of active material (LAM) and ohmic resistance increase (ORI) . LLI is generally the primary source, leading to direct capacity fade ansean_lithium-ion_2019 . So ICA curves retain a certain degree of physical interpretability. Thanks to these compelling properties, during the last ten years SoH - and RUL estimation based on ICA features have become popular. wang_state--health_2023 maps the extracted ICA features simply by a low order polynomial fit to the SoH estimation. They are selecting one or more features depending on charging or discharging cycle and the initial SoC . A further advanced method is suggested in li_state_2020-1 , where the mapping from ICA features to SoH is done with a trained SVR . In both papers, training data is also taken from similar cells, while the test cell is regarded as previously unseen. Also RUL estimation is already done using ICA features, xia_state_2023 apply a bidirectional gated recurrent unit (BiGRU) for RUL estimation. pang_lithium-ion_2021 even use GPR as regression method, both papers train single cell aging models with cycles from BOL to a certain threshold and test for cycles from the threshold to EOL . 1.5 Proposed Solution In a typical application, a battery returns from an arbitrary unknown usage. After completing one (partial) charging cycle, SoH and RUL should be estimated under assessable uncertainties. Based on estimation and uncertainty, a decision on the further use can be taken. A proper method for uncertainty estimation in ML regressors are Gaussian processes (GPs) rasmussen_gaussian_2008 . There are publications like li_state_2020 , also using ICA and GPR for SoH estimation. However, they perform single cell estimation using the first 55% of the cycles as training data to predict further development. In our study, we focus on a scenario where the regression maps one charging cycle of a previously unseen

Read the complete text