School of Engineering Department of Mechanical and Aerospace Engineering 186 Physics Driven Neural Network and Its Application in Fast Cardiovascular Disease Diagnosis Supervisor: HU Wenqi / MAE Student: LUI Hiu Pang / AE Course: UROP 2100, Summer Physics-Informed Neural Networks (PINNs) achieved success in approximating partial differential equations (PDEs) by embedding physical laws into deep learning. However, PINNs can fail to converge due to optimization challenges, especially for nonlinear PDEs like Navier-Stokes. This progress report proposes an enhanced PINN framework extending the Retain-Resample-Release (R3) algorithm with pseudo-labeling for semi-supervised learning and a reinforcement learning inspired replay buffer. Pseudo-labeling provides initial labels for interior points and contributes to model training via exponentially decaying loss. The efficient R3 Sampling uses residual-parameterized Gaussian resampling for high-error focus, with replay buffer reintroducing released collocations to prevent collocation collapse in high residual regions. Code is open-sourced for verification at: https://github.com/VYPang/PINN_pseudoLabeling_replayBuffer Underwater Jellyfish Like Robot for Trash Cleaning Supervisor: HU Wenqi / MAE Student: CHEONG Wei Teng / AE D'SOUZA Pierre Jake / AE HONG Yong Jian / MECH Course: UROP 1000, Summer This paper presents a comprehensive study on the design and optimization of a jellyfish-like underwater robot, focusing on three critical aspects: motor selection for a brace structure, development of a polypropylene flexible hinge, and CFD-based rotor optimization. In this study, CAD software like SOLIDWORKS was used to model the flexible hinge, the movement of scissors robotics arm and impeller. CFD software like ANSYS were utilize to model the fluid flow around the impeller. The motor selection process involved using matrix and vector rotation analysis to determine the power requirements for extending scissor robotics arms under a 10 N load, leading to the selection of the MG996R servo motor with a safety factor of 3. A torsion-spring-inspired polypropylene hinge that has three degrees of freedom that connects the scissor robotics arm and main body of the underwater robot was designed to provide sufficient flexibility when the robotic arm operates underwater. Finally, CFD was employed to optimize an impeller, achieving a 29.77% increase in pressure drop and a 24.1% reduction in back force, significantly enhancing propulsion efficiency. Together, these contributions advance the functionality and performance of bioinspired underwater robotic systems.
RkJQdWJsaXNoZXIy NDk5Njg=