Data-Driven Dynamic Reliability Assessment of Lithium-Ion Battery Considering Degradation Mechanisms

本项目的目标是建立一个锂离子(Li-ion)电池的动态可靠性评估平台。锂离子电池的实时诊断/预测和预测性维护/控制对于电池在从混合动力和电动汽车(HEV/EV)和医疗设备到新兴的智能电网和全电动飞机等广泛的电池供电应用中的可靠和安全运行至关重要。提出的平台使电池管理系统(BMS)能够通过同时分析退化机制和预测故障模式来开发锂离子电池的预测性维护/控制。这项研究的成功将促进我们对如何延长锂离子电池的寿命和防止灾难性故障的理解，并可能导致开发出比现有设备更耐用、更安全的电池供电设备。该项目将通过与领先的行业合作伙伴合作，展示可植入应用的锂离子电池平台，向电池行业传播研究成果。该项目将提供广泛的教育和推广计划，包括1)将研究成果纳入可靠性工程课程，2)利用大学研究计划吸引本科生和K-12学生从事工程工作，3)在主要会议上组织关于锂离子电池失效预防设计的论文会议和小组讨论。到目前为止，锂离子电池的实时诊断/预测还只是经验上的，而且在很大程度上是与潜在的退化机制隔离开来的。这可能归因于缺乏对退化机制和失效模式之间的因果关系的认识。该项目将通过创建一个动态可靠性评估平台，促进基于物理的建模和基于传感器的预测的协同集成，来弥合物理机制和功能故障之间的差距。该平台将允许：1)通过在线估计退化参数来确定和定量分析多种退化机制；以及2)通过在线预测其剩余使用寿命(RUL)来预测失效模式。该平台的创建涉及三项研究工作：1)多物理模型的验证，它使用高精度的充放电循环数据更新多物理电池模型；2)培训健康估计员，它采用机器学习来从一次充电曲线测量中定量地分析多个退化机制；以及3)故障模式预测，它利用定量退化分析来预测故障模式RUL。该平台提供了利用预见性和预见性信息的预测性维护/控制来实现BMS的故障预防能力所需的方法和工具。虽然本项目关注的是使用测量的电气数据动态更新电池可靠性的具体情况，但该方法也适用于其他工程案例，其中使用测量的数据动态更新可靠性估计，以支持维护/控制决策。

The objective of this project is to create a dynamic reliability assessment platform for lithium-ion (Li-ion) battery. Real-time diagnostics/prognostics and predictive maintenance/control of Li-ion battery are essential for reliable and safe battery operation in a wide-range of battery-powered applications, from hybrid and electric vehicles (HEVs/EVs) and medical devices, to the emerging smart grid and all electric airplanes. The proposed platform enables a battery management system (BMS) to develop predictive maintenance/control of a Li-ion battery through concurrently analyzing degradation mechanisms and anticipating failure modes. Successful execution of this research will advance our understanding of how to extend life and prevent catastrophic failure of Li-ion batteries, and will potentially lead to development of battery-powered devices that are more durable and safer than current devices. This project will disseminate research findings to battery industry by demonstrating the platform with a Li-ion battery in an implantable application, through collaboration with a leading industry partner. The project will offer a wide range of education and outreach programs, including 1) incorporating research findings into the Reliability Engineering curriculum, 2) leveraging university research programs to attract undergraduate and K-12 students to engineering career, and 3) organizing paper sessions and panels on Design for Failure Prevention of Li-Ion Battery in major conferences. To date, real-time diagnostics/prognostics of Li-ion battery has been exploited only empirically and largely in isolation with the underlying degradation mechanisms. This may be attributed to the lack of cognizance of the causal relationship between degradation mechanisms and failure modes. This project will bridge the gap between physical mechanisms and functional failures by creating a dynamic reliability assessment platform that facilitates a synergistic integration of physics-based modeling and sensor-based prognostics. The platform will allow for: 1) identification and quantitative analysis of multiple degradation mechanisms through online estimation of the degradation parameters; and 2) anticipation of the failure modes through online prediction of their remaining useful lives (RULs). The creation of the platform involves three research thrusts: 1) validation of multiphysics models, which updates multiphysics battery models using high precision charge-discharge cycling data; 2) training of health estimators, which adopts machine learning to quantitatively analyze multiple degradation mechanisms from a single measurement of charge curve; and 3) prognostics of failure modes, which leverages the quantitative degradation analysis for prediction of failure mode RULs. The platform provides the methods and tools needed to leverage prognostics and prognostics-informed predictive maintenance/control for achieving the failure prevention capability of BMS. Although this project focuses on the specific case of dynamically updating battery reliability with measured electrical data, the methodology will be applicable to other engineering cases in which measured data are used to dynamically update reliability estimates that support maintenance/control decision making.