Quantitative diagnosis of internal short circuit for cylindrical li-ion batteries based on multiclass relevance vector machine

Abstract

Battery internal short circuit (ISC) state should always be supervised. To facilitate the use of battery thermal behaviors, this work develops a lumped thermal evolution model (TEM) based on the equivalent circuit model (ECM). Then, multiple dispersedly-configured TEM/ECM sub-models are synthesized using the extreme learning machine to approximate the distribution nature of real batteries. To acquire ISC dataset, three kinds of active-destruction experiments are carried out on the battery. Thereafter, from thermal and electrical residuals, four ISC features are extracted whereby the multiclass relevance vector machine is utilized to discriminate ISC state. Specially, according to the posterior probability outputs, ISC degree can also be quantified. Experimental results on cylindrical li-ion batteries verify the reliability of the model structure and suggest the proposed diagnosis scheme can recognize ISC faults effectively with a grade misjudgment rate of 14.59% and a state misjudgment rate as low as 3.13%.