Machine learning-based hybrid thermal modeling and diagnostic for lithium-ion battery enabled by embedded sensing

Abstract

Accurate monitoring of internal temperature distribution is critical to the safety of lithium-ion batteries (LIBs). However, both the radial and the axial thermal inhomogeneities are remarkable in practical LIB utilizations, which challenges the control-oriented thermal modeling. Motivated by this, a novel smart battery implanting internally the distributed fibre optical sensor is designed to perceive the inhomogeneity of temperature distribution of LIB. Enabled by this, a hybrid lumped-thermal-neural-network (LTNN) model is proposed, for the first time, by combining the mechanism-driven distributed lumped thermal model and the machine learning-based axial thermal gradient compensation. A hybrid LTNN-based close-loop observer is further proposed to estimate the internal multi-point temperature of LIB in a real-time fashion. Experimental results suggest that the proposed hybrid LTNN model captures the complicated thermal distribution of LIB with remarkably elevated accuracy, compared with the traditional lumped thermal model. Moreover, the hybrid LTNN model is highly compatible with commonly-used state observation methods to realize accurate and space resolved internal thermal diagnostic for the LIB.