Synthetic Data Fusion for Performance Enhancement of Deep Learning Prognostic Models of Lithium-ion Batteries

Lithium-ion batteries are used in various applications, such as energy storage systems and electric vehicles. Estimating the state of health (SOH) and remaining useful life (RUL) of lithium-ion batteries are critical for safe and reliable system operation. Deep learning algorithms display promising performance in predicting SOH and RUL of lithium-ion batteries. However, the performance of these algorithms heavily depends upon the amount of available data. In this article, we propose a novel method to enhance the performance of the deep learning models by fusing synthetic data through transfer learning. The proposed method will reduce the need for data required to train deep learning algorithms, and it can also be applied to existing prognostic models. To validate the effectiveness of the proposed method, we generate synthetic data from two different lithium-ion capacity fade datasets. Results have shown that the proposed method increases the performance of the pre-trained models by 22% and 13% by increasing their diversity through synthetic data.