Mary Lidiya Kalathiparambil Kennedy
PhD
Expert Guided Reinforcement Learning for Decision Support in Collective Control and Maintenance of Wind Farms
Host organization
ETH Zürich
Project Description
The global energy transition driven by sustained efforts against climate change will see wind power double its current electricity generation share. Yet tackling this expansion requires wind power to become more economically feasible and reliable by reducing downtime, maintenance costs, and operational expenditures—ultimately lowering the Levelised Cost of Energy. Addressing this challenge warrants transition to smart infrastructure management paradigms that can be realised by use of intelligent control and automation technologies to make informed, timely and adaptive decisions. Core to such autonomous wind farm operations are rigorous health assessments, well-defined mitigation strategies, and expert-informed criteria addressing system robustness, uncertainty, and operational reliability. In this line of thought, this project implements a graph-based topology-aware reinforcement learning framework that learns from expert feedback for degradation-informed control and maintenance planning. By embedding expert knowledge within data-driven, topology-aware control systems, our framework advances next-generation smart wind energy infrastructure.
​
Supervisors
Eleni Chatzi (ETH Zürich)
Daniel Straub (TU Munich)​
Background
I was born and raised in Kochi, India, where I completed my Bachelor's in Civil Engineering. Following graduation, I had a short research stint in physics-informed machine learning and worked at a startup focused on construction technology. I then moved to Germany to pursue a Master's in Civil Systems Engineering at TU Berlin.
​
Throughout my Master's studies, I have worked on several EU and German nationally funded research projects, including EU H2020 ASHVIN, Urban Water Interfaces, and a Zukunft Bau Research Project, centered on construction digital twins, physics-informed machine learning for CFD modelling, and Bayesian inference-assisted calibration of energy system models, respectively. In my free time, I enjoy dancing, fencing, and reading.
Motivation
Being an engineer with a strong conviction that humanity should benefit from science, I have always been driven towards solving problems that simultaneously push the boundaries of science while directly contributing to technological development. Undoubtedly, our engineering abilities and perception of the universe have gone much further than what we would have developed from empirical knowledge alone—like in fluid flow modelling. Even though we have not been able to fully solve the Navier-Stokes equations, we have applied them to engineer machines from aircrafts to weather prediction models. My passion for fluid flow modelling and machine learning has always driven my research interests. What excites me as an aspiring systems engineer is dissecting complex systems—understanding how they function, identifying their limitations, and creating solutions to make them more efficient and effective. IntelliWind presents an opportunity to work on projects that appeal to both my technical interests and my penchant for tackling system-level problems.