| dc.description.abstract | Lithium-ion (Li-ion) batteries are commonly used in various industrial and domestic
applications, such as portable communication devices, medical equipment, and electric vehicles.
However, the Li-ion battery performance degrades over time due to the aging phenomenon,
which may lead to system performance degradation or even safety issues, especially in vehicle
and industrial applications. Reliable battery health monitoring and prognostics systems are
extremely useful for improving battery performance, to diagnose the battery’s state-of-health
(SOH), and to predict its remaining-useful-life (RUL). In general, it is challenging to accurately
track the battery's nonlinear degradation features as battery degradation parameters are almost
inaccessible to measure using general sensors. In addition, a battery is an electro-chemical
system whose properties vary with variations in environmental and operating conditions.
Although there are some techniques proposed in the literature for battery SOH estimation and
RUL analysis, these techniques have clear limitations in applications, due to reasons such as
lack of proper representation of the posterior probability density functions to capture and model
the nonlinear dynamic system of Li-ion batteries. In addition, these techniques cannot effectively
deal with the time-varying system properties, especially for long-term predictions.
To tackle these problems, a novel hybrid prognostic framework has been developed in
this PhD work for battery SOH monitoring and RUL prediction. It integrates two new models:
the model-based filtering method and the evolving fuzzy rule-based prediction technique. The
strategy is to propose and use more efficient techniques in each module to improve processing,
accuracy and reliability. Firstly, a newly enhanced mutated particle filter technique is proposed
to enhance the performance of particle filter technique and improve the modeling accuracy of the
battery system’s degradation process. It consists of three novel aspects: an enhanced mutation
approach, a selection scheme, and an outlier detection method. Secondly, an adaptive evolving
fuzzy technique is suggested for long-term time series forecasting. It has a novel error-assessment method to control the fuzzy cluster/rule generation process—also, a new optimization
technique to enhance incremental learning and improve modeling efficiency. Finally, a new
hybrid prognostic framework integrates the merits of both proposed techniques to capture the
underlying physics of the battery systems for its SOH estimation, and improve the prognosis of
dynamic system for long-term prediction of Li-ion battery RUL. The effectiveness of the
proposed techniques is verified through simulation tests using some commonly used-benchmark
models and battery databases in this field, such as the one from the National Aeronautics and
Space Administration (NASA) Ames Prognostic Center of Excellence. Test results have shown
that the proposed hybrid prognostics framework can effectively capture the battery SOH
degradation process, and can accurately predict its RUL. | en_US |
