CAREER: Enhancing the State of Health and Performance of Electronics via in-situ Monitoring and Prediction (SHaPE-MaP) - Toward Edge Intelligence in Power Conversion

University Of Houston, Houston TX

Investigators

Abstract

With over 80 % of electricity expected to flow through power converters by 2030, there is a growing requirement to nearly double their operational lifetime (e.g., 50 years for solar PV systems, 30 years for offshore wind, etc.) for reducing the overall carbon footprint. On the other hand, applications like data centers frequently replace their power hardware preemptively to avoid potential outages. The underlying problem with most existing converter installations is the difficulty to assess their health or adapt their performance in real-time without disrupting their operation. To achieve dynamic performance enhancement and reliability improvement, there is a critical need for mission profile-oriented design methods and seamless integration of data-driven prognostic health management with power converters onboard, referred to as ‘Edge Intelligence’. The proposed ‘SHaPE-MaP’ framework aims to enhance the State of Health and performance of Power Electronics via in-situ Monitoring and Prediction using onboard FPGAs or processors for edge intelligence. If successful, the SHaPE-MaP framework will enable the identification of aged or potentially failing modules in real-time and avoid ‘preemptive decommissioning’, thereby increasing the operational life. It will further benefit the system operators in decision-making about maintenance or repair, and the supply chain team to better estimate the logistics or manage the inventory items. Hence, this project will have a widespread impact on most large-scale converter applications, potentially saving hundreds of millions, if not billions of dollars. Moreover, a robust education program will augment this interdisciplinary project to engage K-12 and college students. Predicting system behavior and health has been restricted to the technology design or prototyping phases, and they are hardly implemented in the final products. It is due to the need for additional computing resources, such as a laptop, to execute these techniques – a seemingly impractical scenario, especially in applications with a large number of power converters (e.g., data centers, solar PV farms, etc.). The proposed SHaPE-MaP framework will be a game-changer in enabling such edge intelligence. This project will advance the state-of-the-art power converter hardware and control systems via the following four key thrusts: (i) Development of integrated devices and onboard systems to estimate degradation at the component level; (ii) In-situ implementation of health prediction techniques and fault handling along with the converters using machine learning; (iii) Status estimation and resilient handling of faults and cyber-attacks at converter nodes using edge intelligence; and (iv) Design of coordinated circuits with embedded systems to minimize the computing resources. 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.