SBIR Phase I: Electric Vehicle Useful Life Prediction

SBIR一期项目的广泛影响是使中低收入的购车者能够充分参与交通电气化，确保所有美国人都能从更低的燃料费用、维护成本和更健康的社区环境中受益。目前人们还不太清楚，如今路上行驶的汽车中的电动汽车（EV）电池组是如何随着时间的推移而退化的，以及它们应该如何被评估。未来电池性能的不确定性直接影响了购买二手电动汽车的信心。该项目建议开发一种电池报告，能够通过对电动汽车样本进行物理检查和评估验证的机器学习模型，远程快速诊断和预测电动汽车电池组的健康状态。该项目的规模、范围和商业成果将加速电动汽车的普及。这个小企业创新研究第一阶段项目通过使用一组机器学习模型，解决了对各种插电式电动汽车（pev）的可扩展、可访问和非侵入性健康状态预测的需求。这将通过收集来自不同品牌、型号、年份和气候的前所未有数量的pev的日常观测数据来实现。该数据集将用于在一组关键的远程访问特征上训练机器学习算法，以预测续航里程和电池健康指标。机器学习方法建立在基于树的集成模型和基于物理实验的线性模型的结合之上，并最终扩展到其他更先进的算法。这个项目第一阶段的目标是(1)招募25000 +更多PEV司机超过6个月,(2)展示一组工具准确足以满足消费者需求为短期退化模型使用明白并在路上,(3)验证机器学习决定健康状态估计与实际实验的一个子集车辆,和(4)使用这些实践实验,以更好地理解之间的关系可以远程访问的代理数据和健康状况。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。

The broader impact of this SBIR Phase I project is enabling middle and low income car buyers to fully participate in the electrification of transportation, ensuring that all Americans can benefit from lower fuel bills, maintenance costs, and healthier community environments. It is not well understood how electric vehicle (EV) battery packs in cars on the road today degrade over time and how they should be valued. This uncertainty in future battery performance directly impacts confidence in buying a used EV. The project proposes the development of a battery report that is able to remotely and rapidly diagnose and predict the state of health of EV battery packs through machine learning models validated by physical inspection and evaluation of a sample of EVs. The project’s scale, scope, and commercial result would accelerate the adoption and accessibility of EVs. This Small Business Innovation Research Phase I project addresses the need for scalable, accessible and non-invasive state of health prediction on a wide variety of plug-in electric vehicles (PEVs) by using a set of machine learning models. This will be accomplished by collecting daily observation data from an unprecedented number of PEVs on the road from a diverse set of makes, models, years, and climates. This dataset will be used to train machine learning algorithms on a key set of remotely accessible features to predict range and battery health metrics. The machine learning approach builds on a combination of tree-based ensemble models and linear models informed by physics-based experiments, and will eventually expand to other more advanced algorithms as appropriate. The goals of this Phase I project are to (1) recruit 25,000+ more PEV drivers over 6 months, (2) demonstrate a set of tools that are accurate enough to meet consumer demand for near-term degradation models for the used PEVs on the road, (3) validate machine learning determined state of health estimates with hands-on experimentation with a subset of vehicles, and (4) use these hands-on experiments to better understand the relationships between remotely accessible proxy data and state of health.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.