Remaining useful life estimations applied on the sizing and the prognosis of lithium ion battery energy storage systems

Abstract

The present thesis develops an accurate sizing tool for the most relevant lithium ion battery energy storage system applications considering the aging and the remaining useful life. The developed tool involves firstly, the construction of the aging models of the lithium ion battery health indicators; secondly, the calculation of the end of life based on the evolution of the modelled health indicators; thirdly, the calculation of the levelized cost of the most relevant applications of lithium ion battery energy storage systems; and fourthly, the minimization of the committed error with the constructed aging models supported by electrode level data and prognosis algorithms. The methodology behind the construction and calculation of all the elements integrated on the sizing tool is described throughout the chapters of this thesis. Firstly, the end of life state of the battery is determined as a combined threshold of all the health indicators of interest. Its calculation requires the implementation of an electro-thermal model in a simulation environment defined by the end of life criteria specified by the application requirements. Secondly, the evolution of health indicators of interest are modelled based on the most relevant stress factors. The methodology to acquire the aging data and the construction of the posterior empirical models are presented. The validation of the constructed models based on the acquired data is performed based on three aspects: the accuracy describing the observed cases, the correctness of interpolations and the real life applicability. Thirdly, the simulation environments for lithium ion battery energy storage systems applied on an electric vehicle application and on a stationary application are developed where the levelized cost of different battery solution sizes is calculated. The simulation environment integrates the already developed electric-thermal model, end of life map and aging models. Fourthly, the error done by the constructed aging models is minimized by focusing on the errors done when extrapolating in time and when facing odd events. On one hand, electrode level data is analysed to generate data artificially and reduce the errors when extrapolating in time. On the other hand, a prognosis stochastic algorithm is selected and employed with real life data to deal with the effect that odd events have on the evolution of the health indicators. The validity of many assumptions made for the development of the end of life map, the aging models, the simulation environment used on the sizing tool, the artificial data generator and the real time prognosis tool are proved experimentally.