RII Track-4:NSF: Spatiotemporal Modeling of Lithium-ion Battery Packs for Electric Vehicle Battery Management Systems

最近关于电动汽车锂离子电池过热和起火的报道引发了人们对用户安全以及电动汽车的广泛接受度的担忧。这些事故凸显了监测和控制电池组的车载电子系统（即电池管理系统（BMS））在检测此类异常行为方面的局限性。因此，为了防止灾难性故障，增强BMS识别电池行为的能力变得势在必行。智能BMS能够实时监控电池组的最小部分，并学习异常行为以进行未来预测，这可能是解决这些安全问题的关键。通过NSF EPSCoR RII track4奖学金项目，PI将与桑迪亚国家实验室（SNL）的专家合作，开发一种革命性的解决方案，用于捕获和学习电动汽车锂离子电池组的动态行为。这种创新方法有望增强BMS的预测能力，并推动以健康为中心的决策。此外，该倡议还包括一个全面的教育和推广部分，旨在促进代表性不足的学生参与研究，将研究成果整合到研究生和本科教育中，并通过在线视频教程促进K-12关于锂离子电池操作和安全的推广。这项研究基础设施改善轨道4 EPSCoR研究人员项目将为阿拉巴马大学亨茨维尔分校的一名助理教授提供奖学金，并为一名研究生提供培训。这项工作将与桑迪亚国家实验室（SNL）的研究人员合作进行。该奖学金项目的主要目标是开发：1)锂离子电池组的互联模型；2)学习空间和时间动态的深度神经网络模型。该项目的内在科学价值在于理解锂离子电池组的电、热、老化行为之间的相互作用，以及这些错综复杂的相互关联的行为如何在空间和时间上影响电池内部降解的传播。利用这些见解，该项目将在目标1中为电池组构想一个相互连接的电热老化模型。还将描述以数据为中心的识别策略，以利用图论和网络推理来估计互联模型的参数。在目标2中，将设计一个深度扩散卷积神经网络（DD-CRNN）来学习群体的时空动态。这种物理驱动的DD-CRNN模型将使用混合实验和合成数据进行训练。依靠SNL扩展的包级测试基础设施，该项目将积累退化和滥用数据，这对于训练DD-CRNN和确认模型的有效性至关重要。所提出的模型和学习框架将通过提供精确的充电状态（SOC）、健康状态（SOH）和热参数估计来改变电池健康监测。这一创新将通过促进单元和模块级别的健康和异常检测，赋予BMS更大的决策自主权。该项目将开辟电源和能源管理的新领域，最大限度地降低电池组过热和火灾事故的风险，确保更安全、更高效地利用电池。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。

Recent reports of lithium-ion (Li-ion) battery overheating and catching fire in electric vehicles (EVs) have raised concerns about user safety and the broader acceptance of EVs. These incidents highlight the limitations of the onboard electronic system that monitors and controls the battery pack, referred to as the battery management system (BMS), in detecting such abnormal behavior. Therefore, enhancing the BMS's capabilities to discern the battery's behavior becomes imperative to prevent catastrophic failures. A smart BMS capable of monitoring the smallest part of a battery pack in real-time and learning abnormal behavior for future prediction could be the key to addressing these safety concerns. Through this NSF EPSCoR RII Track-4 fellowship project, the PI will collaborate with experts at the Sandia National Laboratory (SNL) to develop a transformative solution for capturing and learning the dynamic behavior of Li-ion battery packs in EVs. This innovative approach promises to enhance the BMS's predictive capabilities and drive health-centric decisions. Additionally, this initiative includes a comprehensive educational and outreach segment, aimed at promoting the participation of underrepresented students in research, integrating research findings into both graduate and undergraduate education, and facilitating K-12 outreach on Li-ion battery operation and safety through online video tutorials.This Research Infrastructure Improvement Track-4 EPSCoR Research Fellows project will provide a fellowship to an Assistant Professor and training for a graduate student at the University of Alabama Huntsville. This work would be conducted in collaboration with researchers at the Sandia National Laboratory (SNL). The primary goals of the fellowship project are to develop: 1) an interconnected model of a Li-ion battery pack and 2) a deep neural network model to learn the spatial and temporal dynamics. The project's intrinsic scientific merit revolves around comprehending the interplay between the electrical, thermal, and aging behavior of the Li-ion battery pack and how these intricately linked behaviors influence internal degradation propagation among cells, both spatially and temporally. Leveraging these insights, the project will, in Aim 1, conceive an interconnected electro-thermal-aging model for the battery pack. A data-centric identification strategy will also be delineated to estimate the parameters of the interconnected model, drawing on graph theory and network inference. In Aim 2, a deep diffusion convolutional neural network (DD-CRNN) will be designed to learn the spatiotemporal dynamics of the pack. This physics-driven DD-CRNN model will be trained using a blend of experimental and synthetic data. Relying on SNL's expansive pack-level testing infrastructure, the project will accumulate degradation and abuse data, which is essential for training the DD-CRNN and affirming the model's validity. The proposed model and learning framework are poised to transform battery health monitoring by delivering precise State of Charge (SOC), State of Health (SOH), and thermal parameter estimations. This innovation will empower BMS with greater autonomy in decision-making by facilitating cell- and module-level health and anomaly detection. The project will chart a new frontier in power and energy management and critically minimize the risk of pack overheating and fire incidents, ensuring safer and more efficient battery utilization.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.