TR2018-152

Remaining Useful Life Estimation of Batteries using Dirichlet Process with Variational Bayes Inference



Rechargeable batteries supply numerous devices with electric power and are critical part in a variety of applications. While estimation of battery's state of charge (SoC), state of health (SoH) and state of power (SoP) have been in research focus in the past years, prediction of battery degradation has recently started to gain interest. An accurate prediction of the remaining number of charge and discharge cycles a battery can undergo before it can no longer hold charge and is declared dead, is directly related to making timely decision as to when a battery should be replaced so that power interruption of the system it supplies power to is avoided. A methodology for inferring probability distribution of the remaining number of charge-discharge cycles of a battery, based on training dataset containing measured discharge voltage waveforms of one or more batteries of similar type, is presented in this paper. The methodology strongly draws on modeling discharge voltage waveforms using Dirichlet Process Mixture Model framework and performs approximate inference using variational Bayes approach. The experimental results corroborate that the proposed method is able to provide useful predictions of the remaining useful life of a battery in early stages of its life.