A Computationally Informed Realisation Algorithm for Lithium-Ion Batteries Implemented with LiiBRA.jl

TL;DR

This paper introduces an open-source, computationally efficient realisation algorithm for lithium-ion batteries, enabling real-time physics-based modeling suitable for in-vehicle applications, validated through experimental testing and performance benchmarking.

Contribution

It presents a novel computationally informed discrete realisation algorithm (CI-DRA) implemented in Julia, reducing model creation time and enabling in-vehicle battery modeling on ARM architectures.

Findings

88% reduction in computational time compared to traditional methods

Successful validation with WLTP testing for LG Chem M50 batteries

ARM implementation achieves models within 5.5 seconds with 43% performance loss

Abstract

Real-time battery modelling advancements have quickly become a requirement as the adoption of battery electric vehicles (BEVs) has rapidly increased. In this paper an open-source, improved discrete realisation algorithm, implemented in Julia for creation and simulation of reduced-order, real-time capable physics-based models is presented. This work reduces the Doyle-Fuller-Newman electrochemical model into continuous-form transfer functions and introduces a computationally informed discrete realisation algorithm (CI-DRA) to generate the reduced-order models. Further improvements in conventional offline model creation are obtained as well as achieving in-vehicle capable model creation for ARM based computing architectures. Furthermore, a sensitivity analysis on the resultant computational time is completed as well as experimental validation of a worldwide harmonised light vehicle test procedure (WLTP) for a LG Chem. M50 21700 parameterisation. A performance comparison to the conventional Matlab implemented discrete realisation algorithm (DRA) is completed showcasing a mean computational time improvement of 88%. Finally, an ARM based compilation is investigated for in-vehicle model generation and shows a modest performance reduction of 43% when compared to the x86 implementation while still generating accurate models within 5.5 seconds.