Predicting unseen lithium-ion battery degradation modes through domain generalization

The rapid development of modern society and the increasing energy demand have made secondary batteries one of the most promising energy storage systems. Among these, lithium-ion batteries are currently the most widely used. Ensuring the reliability of these lithium-ion battery systems necessitates accurate diagnosis and prediction of battery degradation modes such as loss of lithium inventory, loss of active material in the negative electrode, and loss of active material in the positive electrode. However, it is often practically challenging to secure training data for all batteries for which degradation mode predictions are required. This paper proposes a model to overcome these limitations, enabling the prediction of battery degradation modes even when the domain of the training data differs from that of the test data. The proposed model is designed to maintain high performance across various domains by integrating convolutional neural networks, domain adversarial neural networks, and MixStyle. Additionally, the LeakyReLU activation function was introduced to enhance domain generalization performance. Experimental results show that the proposed model exhibits excellent domain generalization performance across various domains. In addition, a real-world case study using a public LG M50T aging dataset confirms that the proposed approach remains effective under practical conditions, where the domain shift is defined by operating temperature and the evaluation is conducted at RPT checkpoints. The superiority of the proposed model was validated through comparative experiments with other models using the Hawai’i Natural Energy Institute diagnostic dataset.