Integrated framework for SOH estimation of lithium-ion batteries using multiphysics features

Seho Son, Siheon Jeong, Eunji Kwak, Jun hyeong Kim, Ki Yong Oh

Research output: Contribution to journalArticlepeer-review

1 Scopus citations


This study proposes a highly reliable, robust, and accurate integrated framework to estimate the state-of-health (SOH) of lithium-ion batteries (LIBs), focusing on feature extraction and manipulation. This framework comprises three phases: feature extraction, feature manipulation, and SOH estimation. First, multiphysics features are extracted from mechanical and electrochemical evolutionary responses as distinct health indicators (HIs) to account for the multiphysics degradation mechanisms. Second, these features are manipulated to eliminate outliers and noises. This phase is especially effective for impedance HIs, considering the high sensitivity of these HIs to minor environmental perturbations. Third, a multivariate Gaussian distribution theory estimates the SOH combined with a nonlinear quadratic kernel to account for nonlinear characteristics in degradation modes of LIBs. The estimated results under various environments verify that the multiphysics feature primarily increases accuracy, whereas the feature manipulation ensures reliability and robustness. However, both phases are complementary in securing the accuracy, reliability, and robustness of the framework. Although the lifespan of LIBs is estimated using the training set in the 5 % SOH range, the estimation errors of the proposed framework are less than 2.5 % in all test sets. Thus, the proposed method ensures its potential applicability in practical implementations of onboard battery management systems.

Original languageEnglish
Article number121712
StatePublished - 2022 Jan 1
Externally publishedYes


  • Denoising autoencoder
  • Gaussian process regression
  • Lithium-ion battery
  • Machine learning
  • State-of-health


Dive into the research topics of 'Integrated framework for SOH estimation of lithium-ion batteries using multiphysics features'. Together they form a unique fingerprint.

Cite this