Examenspresentation av Anton Albinsson och Thomas Trinh

Examiner: Torsten Wik

Abstract:
The increasing adoption of electric vehicles (EVs) in recent years has put an increasing
focus on the essential component in these vehicles: lithium-ion batteries.
Accurately predicting the End-of-Life (EoL) of batteries is crucial for automotive
manufacturers, such as Volvo Cars, to ensure optimal performance and operational
safety of their fleet of EVs.
Predictions of EoL pose several challenges as empirical observations of EoL from
EVs are limited, while the experiments in laboratory settings cannot fully replicate
the operating conditions in a real-world setting. Numerous methods exist today to
predict EoL on EVs based on data from either a fleet of EVs or laboratory experiments.
This thesis explores a framework for predicting the EoL of a fleet of EVs by combining
lab and fleet observation data using a network-based transfer learning model.
The framework’s purpose is to transfer degradation information from the lab data
to real-world fleet data to predict EoL more accurately. As part of the framework,
an ANN is first pre-trained on a dataset and then fine-tuned on a second dataset
where most layers have their parameters frozen.
To evaluate the framework’s effectiveness, a two-part experimental study is conducted
using datasets with degradation paths to EoL. The first part evaluates how
much the fine-tuning amount affects predictive performance, and the second part
investigates the knowledge transfer between the pre-trained and fine-tuned model.
The experimental results demonstrate that the predictive performance improves
with an adequate amount of fine-tuning data. Furthermore, fine-tuning the model
also shifts the predicted degradation path closer toward the ground truth, indicating
that knowledge has successfully been transferred during the fine-tuning process.
Furthermore, applying the framework with fine-tuning on a fleet of EVs, the model
demonstrates reasonable degradation paths down to the EoL for the fleet, highlighting
its ability to capture complex degradation mechanisms of batteries.

Welcome!

Anton, Thomas and Torsten