Degradation-Aware Self-Healing Control of Power Electronics Systems

计算能力无处不在。传感器的成本越来越低，而且无处不在。尽管资源丰富，但现代电力电子系统（PESs）无法精确定位其退化状态，因此无法进行自我修复以防止代价高昂的故障。例如，风力涡轮机或光伏（PV）系统遭受极端温度和湿度的波动，从-30ºC到55ºC和30%到100%（例如，海上应用）。如此恶劣的气候和热（C&amp；T）波动迅速增加了故障率和维护成本，高达总发电成本的30%。假设它们的退化和剩余使用寿命（RUL）可以被精确地测量或提前精确地预测。在这种情况下，我们可以利用现有的PESs软件或硬件，通过对日益复杂的现代能源系统中电力电子构件的退化演变、积累、加速以及RUL变化的自适应控制，来执行主动的自我修复。这可以大大提高可靠性、调度灵活性和可控性，同时防止代价高昂的停机。该项目的成果将用于互动和实践学习计划，以激发K-12儿童对STEM领域的兴趣。该项目将模拟现实世界中C&amp；T波动下宽带隙（WB）功率开关的退化，这是现代PESs中开发退化感知自愈（DASH）控制的关键瓶颈。具体来说，我们将开发一种级联生成对抗网络学习和数据净化策略，以有效地模拟现实世界C&amp；T条件下电力电子设备的大量可靠性数据。制定的数据驱动模型和多感官工具将从根本上比最先进的技术更准确。此外，我们将开发一个系统的DASH控制框架，通过理解四种系统健康状况（健康、中度退化、自我修复和故障），而不是传统的启发式假设（健康和故障），实现终身管理的PES操作。制定的RUL估计和DASH控制工具能够从根本上改变当前的设计和控制实践，将可靠的WB开关无缝集成到各种C&amp条件下的电力电子和能源系统中。这将加速向节能电网和交通电气化的过渡，同时最大限度地降低开发成本和周期，防止意外停机和灾难性故障。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。

Computational power is everywhere. Sensors are increasingly low-cost and ubiquitous. Despite the extensive resources, modern power electronics systems (PESs) cannot pinpoint its degradation status and, hence, cannot perform self-healing to prevent costly failures. For example, wind turbines or photovoltaic (PV) systems are subjected to extreme temperature and humidity swings from -30ºC to 55ºC and 30% to 100% (e.g., offshore applications). Such a harsh climate and thermal (C&T) swings rapidly increase the failure rate and maintenance costs by up to 30% of the overall generation cost. Suppose their degradation and, hence, remaining useful lifetime (RUL) can be accurately measured or precisely predicted in advance. In that case, we can utilize existing PESs software or hardware to perform proactive self-healing through the adaptive control of degradation evolution, accumulation, acceleration, and, hence, RUL changes of the building blocks of power electronics in increasingly complicated modern energy systems. This could substantially enhance reliability, scheduling flexibility, and controllability while preventing costly downtimes. The outcome of this project will be utilized for interactive and hands-on learning programs to inspire K-12 children’s interest in STEM fields.This project will model the degradation of wide bandgap (WB) power switches under real-world C&T swings, which poses the critical bottleneck of exploiting degradation-aware self-healing (DASH) control in modern PESs. Specifically, we will develop a cascade generative adversarial networks learning and data purification strategy to effectively model the large reliability data of power electronics under a real-world C&T condition. The formulated data-driven models and multi-sensory tools will be fundamentally more accurate than state-of-the-art. Moreover, we will develop a systematic DASH control framework, enabling lifetime managed PES operations by understanding four system health conditions (healthy, intermediate degradation, self-healing, and failure) instead of a traditional heuristic assumption (healthy and failure). The formulated RUL estimation and DASH control tools are able to fundamentally transform the current design and control practices, creating a seamless integration of reliable WB switches into the wide spectrum of power electronics and energy systems under diverse C&T conditions. This will accelerate the migration toward an energy-efficient grid and transportation electrification while minimizing development cost and period and preventing unplanned downtime and catastrophic failures.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.