부산대학교 도서관

Determining the age of a battery under use is a highly critical but unsolved problem. On one hand , knowing battery age improves the reliability of the system using the battery. On the other hand, safety critical systems have to use new batteries on the start of each run, resulting in huge battery costs. There are numerous attempts at finding the battery age but owing to the nonlinear nature of the chemical processes in a battery, its modeling is a challenging problem. Battery ageing problem has been traditionally treated in a model-based FDI framework. The FDI techniques mostly depend upon the precision of the battery model. Obtaining accurate battery models is still an unsolved problem. Recently, data driven methods are getting acceptable by the industry owing to the ready availability of data and the growing trend towards data analytics. However, these data driven methods are not able to benefit from the battery dynamics and ground truth, considering the battery as a black box. A hybrid but rigorous approach, based on structural analysis and clustering concepts, is proposed which takes the best of both worlds by devising data driven FDI methodology benefiting from the ground truth provided by the battery models based FDI. This approach puts rigour into the classical data driven approaches. It also earmarks the right battery variables to be used in the data analysis, thus considerably decreasing the data size involved. Also it annotates the results of unsupervised learning which are hard to understand. The proposed hybrid approach is applied on predicting the age of Lithium-Ion battery showing the efficacy of the approach while demonstrating the effective use of the battery data with an approximate model of the battery.