Applied Physics Seminar
***Refreshments at 3:45pm outside Noyes 147
Abstract:
Plasma is a ubiquitous phenomenon in both space and industry. To measure the state of plasma, microparticles provide a minimally intrusive alternative to traditional probes, as their motion inherently encodes information about the surrounding plasma environment. In this work, we demonstrate how known physical laws can be integrated with machine learning (ML) to extract this information. Using 3D trajectories of approximately 20 non-identical microparticles, our model simultaneously predicts particle interaction forces (dependent on position and size), environmental forces, and damping coefficients, fitting the sum of these forces to particles' accelerations with a cross-validation R2>0.99. These predictions can be further analyzed using physical laws to infer plasma characteristics, such as particle charges and the electric field. Remarkably, aside from experimental trajectories, no explicit physical laws were used to train the model; yet, the model independently uncovered fundamental physical principles, including (a) the electric field is independent of the mass of the test particle and (b) the electric field is curl-free in the absence of a changing magnetic field. This confirms that the model accurately captures the underlying physics of the system, effectively transforming microparticles into reliable probes for plasma diagnostics.
More about the Speaker:
Wentao Yu earned his B.S. degree in Physics from Zhejiang University in 2018, where he conducted research in hard condensed matter. He subsequently joined Emory University to pursue a Ph.D. in Physics, which he is on track to complete by December 2024. During his time at Emory, Wentao participated in soft condensed matter and plasma physics labs, mentored two undergraduate students on their honors theses, served as a teaching assistant for four courses, and actively participated in outreach initiatives such as science clubs that introduced engaging physical experiments to primary school students.
In his doctoral research, he created a tomographic setup capable of tracking the 3D trajectories of individual microparticles in plasma. The COVID-19 pandemic temporarily shut down his lab, but this unexpected pause offered him the opportunity to delve into artificial intelligence (AI) in the old era when there was no GPT. Post-pandemic, Wentao transitioned from purely experimental physics to an experimental-AI hybrid approach. Recognizing the potential of AI to analyze large-scale experimental datasets—including 3D trajectories of approximately 20 microparticles across 10,000 frames—he developed physics-informed AI models to uncover the underlying mechanisms driving these trajectories. His current research focuses on integrating AI with experimental physics, and he is excited to share insights from this work.