Fault detection of a Li-ion battery using SVM based machine learning and unscented Kalman filter. (2023)

• Record Type: Journal Article
• Authors: Chatterjee, Sayanti; Kumar Gatla, Ranjith; Sinha, Pampa; Jena, Chitralekha; Kundu, Shubhasri; Panda, Babita; Nanda, Lipika; Pradhan, Arjyadhara
• Is Part Of: Materials today. Volume 74(2023)Part 4
• Page Start: 703
• Page End: 707
• Publication Date: 2023
• Subjects: State of Charge -- UKF -- Fault Detection -- SVM -- Estimation; Materials science -- Congresses -- Periodicals; 620.1
• DOI: 10.1016/j.matpr.2022.10.279
• Languages: English
• ISSNs: 2214-7853

Abstract

Abstract: The focal aim of the current paper is to present a unique support vector machine (SVM) based fault diagnosis technique for a Lithium ion battery. This paper basically objectifies the derivative free estimator paradigm for the assessment of State of Charge (SOC) for a Lithium-ion (Li-ion) battery. The estimation error has been taken as the references for support vector machine (SVM) based learning for fault detection. The model of Li-ion battery which is instrumented here is basically the lumped model. One of the primary drawbacks of this type of battery is the estimation of the state of charge (SOC) perfectly because of the electrochemical topographies. These features alters with alteration of physical constraints which causes the increase of process and measurement noise levels while updating the online battery model. These drawbacks also create some fault condition in battery management system. The short circuit current due to abrupt change in state of charge cause short circuit fault which is not desirable. Unscented Kalman Filter obviates the noise level by minimizing estimation error and fault can be diagnosed accurately using machine learning with great accuracy and less number of false alarms. In one word the main novelty of this paper is to detect fault of state of charge of Li-ion battery using SVM based machine learning where the vector machine is trained by the error value deduced by the UKF. It is proven from the simulation studies that machine learning can detect fault faster with higher range coverage than other statistical methods.