Session: Research Posters

Paper Number: 166566

Data-Driven Modeling of Collision Dynamics in Trapezoidal Electrode Mems Energy Harvesters 

As modern devices continue to shrink in scale, the demand for energy sources of comparable size grows in tandem. At the same time, a global shift toward renewable solutions has highlighted ambient energy scavenging as the preferred solution for microscale sensors. Harnessing the often-overlooked by-product of vibration energy offers a promising route for powering such devices. Among the various conversion approaches, electrostatic vibrational energy harvesters (E-VEHs) stand out due to their compatibility with microfabrication processes and their potential for on-chip integration with sensing units.

While E-VEHs may appear simplistic in design, their underlying behavior can be remarkably complex, necessitating rigorous analytical models to capture system dynamics. More complex designs may also feature collision, inducing a phenomenon known as frequency-up conversion, a method to improve energy generation per cycle. These complex models are valuable when designing harvester geometries and evaluating performance but often have notable differences compared to real-world devices. In particular, device ringing and electrode collision dynamics are not fully captured in current modeling efforts, a problem addressed in this work. Moreover, geometric imperfections and nonideal manufacturing conditions can introduce additional sources of dissipation and variability, further complicating the collision-driven energy harvesting process. As a result, purely analytical methods struggle to capture the high-amplitude transient effects or subtle nonlinearities observed in experiments, necessitating more comprehensive modeling approaches.

The paper presented here aims to address the gap between experimental observations and existing analytical models by leveraging a Universal Differential Equation (UDE). Evolving from the concept of neural ordinary differential equations, a UDE embeds a first-principles framework and augments it with correction terms parameterized by a neural network. This hybrid approach not only refines specific analytical terms but provides insight into the nature of model errors. In doing so, the model provides context for previously understood nonlinearities and allows for additional refinement of their terms. In preliminary tests, the UDE-based model shows significantly improved agreement with experimental data, particularly in capturing the high-amplitude ringing and electrode collision dynamics. Moreover, these initial results confirm the model’s ability to accurately reproduce frequency-up conversion, promising more reliable performance predictions across diverse operating conditions. Furthermore, better understanding can be made to pull-in, a phenomenon that constrains the design of devices and inhibits certain conditions ideal for energy harvesting. By integrating both theoretical and data-driven corrections, the UDE framework yields high-fidelity predictions that guide the design and optimization of next-generation MEMs harvesters, ensuring real-world performance matches model outcomes.

Presenting Author: Matthew Galarza Rensselaer Polytechnic Institute

Presenting Author Biography: I am a doctoral candidate in Mechanical Engineering at Rensselaer Polytechnic Institute, specializing in nonlinear dynamical systems, nonlinear control theory, and scientific machine learning. My research focuses on coupled electromechanical energy harvesters, examined through both theoretical and experimental lenses.

Authors:

Matthew Galarza Rensselaer Polytechnic Institute
Shaowu Pan Rensselaer Polytechnic Institute
Diana-Andra Borca-Tasciuc Rensselaer Polytechnic Institute