Battery capacity estimation using 10-second relaxation voltage and a convolutional neural network

Accurate and reliable battery capacity estimation has been a key challenge to technology advancement of safety-critical systems such as electric vehicles and stationary energy storage systems. In this work, we propose a new battery capacity estimation approach using relaxation voltage data collected for only 10 s. A strong correlation is first identified between the relaxation voltage and battery capacity over the entire lifetime. Based on this key enabling correlation, a convolutional neural network model is then developed to estimate capacity for batteries with different degradation paths. The generalizability of the model is assessed by 28 batteries and the average percentage test error on 8 validation cells is only 1.8%. The high predictive power of the relaxation voltage is rationalized by demonstrating the impacts of degradation mechanisms on the ionic and electronic transport properties as the battery ages, ultimately manifesting in the relaxation voltage curves. We also show that the proposed method has the potential to be extended to batteries with different chemistries.

Automated summary

In this work, a new battery capacity estimation approach using relaxation voltage data is proposed. A correlation is identified between the relaxation voltage and battery capacity, and a convolutional neural network model is developed to estimate capacity for batteries with different degradation paths. The model shows high predictive power with a low average test error of 1.8%. The method also has the potential to work for batteries with different chemistries.