Real-Time Prognostics of Lithium-ion Batteries in Electric Unmanned Aerial Vehicles

由于能量和功率密度高、电压容量大、重量轻，锂离子电池正成为电动无人机的主要动力推进来源。电池性能通常由于电化学反应而下降，也称为电池老化。电池老化可能导致容量衰减甚至灾难性故障。因此，电池的实时健康管理对电动无人机的安全性和可靠性至关重要。现有电池健康管理技术的局限性包括：(1)很少有电池健康管理技术能够实现对放电容量和放电结束的实时预测，因为现有技术需要充电周期的状态监测数据，而这些数据在飞行过程中并不总是可用的；(2)目前的电池健康管理技术仅在简单、固定的飞行计划和恒定的有效载荷下有效。有效的实时电池健康管理技术不仅会对航空航天，还会对医疗保健、汽车、国防和物流行业产生重大影响，在这些行业中，电池有着广泛的应用。该研究的目标是开发一种新型电池健康管理技术，能够实时预测电动无人机中锂离子电池的放电容量和放电结束。具体而言，将开发一种结合两种深度学习算法的新型计算框架。一种深度学习算法通过提取放电周期之间的空间相关性来预测放电容量；另一个通过捕获放电周期内的时间依赖性来预测放电结束。此外，将开发一种基于最优传输的领域自适应技术，通过跨飞行计划和有效载荷的知识转移来预测不同飞行计划和有效载荷下的放电能力和放电结束。提出的实时电池健康管理技术将使用从电动无人机收集的实验数据进行验证。本研究将通过回答以下研究问题，推动电池健康管理领域的发展：(1)仅在飞行过程中，利用在放电周期中采集的状态监测数据，能否实时预测锂离子电池的放电容量和放电终值？(2)一个飞行计划和有效载荷下锂离子电池的放电容量和放电终点是否可以用于预测另一个飞行计划和有效载荷下的放电容量和放电终点？该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Due to high energy and power density, high voltage capacity, and light weight, Lithium-ion batteries are becoming the primary source of power propulsion for electric unmanned aerial vehicles. Battery performance usually degrade due to electrochemical reactions, also known as battery aging. Battery aging could result in capacity fade and even catastrophic failure. Therefore, real-time battery health management is crucial to the safety and reliability of electric unmanned aerial vehicles. The limitations of existing battery health management techniques include (1) few battery health management techniques can achieve real-time prediction of discharge capacity and end-of-discharge because existing techniques require condition monitoring data in charge cycles, which are not always available during flight; (2) current battery health management techniques are effective only under simple, fixed flight plans and constant payloads. Effective real-time battery health management techniques will make a significant impact on not only the aerospace but also healthcare, automotive, defense, and logistics industries where batteries have a wide range of applications. The objective of the proposed research is to develop a novel battery health management technique that enables real-time prediction of discharge capacity and end-of-discharge of Lithium-ion batteries in electric unmanned aerial vehicles. Specifically, a novel computational framework that combines two deep learning algorithms will be developed. One deep learning algorithm predicts discharge capacity by extracting spatial correlations between discharge cycles; the other predicts end-of-discharge by capturing temporal dependencies within a discharge cycle. In addition, an optimal transport-based domain adaptation technique will be developed to predict discharge capacity and end-of-discharge under varying flight plans and payloads through the transfer of knowledge across flight plans and payloads. The proposed real-time battery health management technique will be validated using experimental data collected from electric unmanned aerial vehicles. The proposed research will advance the field of battery health management by answering the following research questions: (1) Can discharge capacity and end-of-discharge of Lithium-ion batteries be predicted in real-time using condition monitoring data collected in discharge cycles only during flight? (2) Can knowledge gained on discharge capacity and end-of-discharge of Lithium-ion batteries under one flight plan and payload be used to predict discharge capacity and end-of-discharge under another flight plan and payload?This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.