A collaboration between the Schools of Mathematics, Engineering and Dukosi has resulted in a paper: "Identification and machine learning prediction of knee-point and knee-onset in capacity degradation curves of lithium-ion cells"
A collaboration between the Schools of Mathematics, Engineering and a local company Dukosi (https://www.dukosi.com/) has resulted in a paper, titled 'Identification and machine learning prediction of knee-point and knee-onset in capacity degradation curves of lithium-ion cells', to appear in Energy and AI journal (https://www.journals.elsevier.com/energy-and-AI). The collaboration consisted of Dr Gonalo dos Reis, Prof. Miguel Anjos and Paula Fermin from the School of Mathematics and Dr Encarni Medina-Lopez from the School of Engineering.
The article came out of a project investigating two-phase degradation behaviour of Li-ion battery cells. A cells' energy storage capacity degrades initially at a low rate and then goes through an accelerated degradation, displaying a so-called knee pattern, until the death of the battery. The occurrence of the knee is a crucial factor of the cycle life of the cell, with important implications for predictive maintenance and technology development.
In this project, people from mathematics and data engineering built an innovative new concept for battery management systems called knee-onset, which gives a very early warning of rapid cell degradation. Mathematical models were developed to identify knee points from capacity degradation curves and to predict when it will occur from a very early stage of the cell's life. The methods used have significant positive commercial applications for the energy storage sector. They are also of considerable academic interest, given the growing attention being devoted to the modelling of cell behaviour.
When tested on a dataset of Li-ion cells with lifespans ranging from 500 to 2000 charge-discharge cycles, the models predict the useful lifespan of a cell with circa 9% error using only the first 50 cycles, and they could grade cells as "short-range", "medium-range" or "long-range" with nearly 90% accuracy using as few as 3 cycles.
The paper can be read in full here: https://doi.org/10.1016/j.egyai.2020.100006