EDUCATION
RESEARCH INTERESTS
Using Multiple Predictions to Infer Response of Mechanical Systems and Reduce Uncertainty
2014 - Present
University of Florida
PhD at Mechanical Engineering
M.S. in Statistics
Methods to Avoid Numerical Instabilities of Topology Optimization
Metal Free Hydrogen ActivationReal-Time Error Compensation Techniques for CNC Machining System
2012 - 2014
University of Florida
M.S in Mechanical Engineering
2008 - 2012
Shanghai Jiao Tong University
B.S. in Mechanical Engineering
MY RESEARCH PROJECTS
Developing a stochastically verified structural toolset using surrogates
-
Compared distribution models to characterize experimental variability of composite laminate and selected Gaussian distribution for further numerical analysis based on hypothesis tests
-
Examined model correction and calibration approaches to approximate experimental strength by incorporating simulations. We found simulation only contributed to avoid large prediction error for scarce experiments. Reasonable surrogates could be identified using cross-validation.
-
Introduced bootstrapping to predict B-basis allowables of composite laminates using surrogates. Estimating B-basis allowables risks large uncertainty and introducing bootstrapping enables conservative prediction with small prediction margin.
-
Worked integratedly (on data analytics) with Materials Science Corporation (on simulation) and National Institute for Aviation Research (on experiments) through remote collaboration
Forecasting response of mechanical systems towards inaccessible domain
-
Proposed a framework for predicting inaccessible point by developing surrogates along lines converging at the target point from inaccessible domain. Multiple predictions are then obtained through 1-D surrogates along each line and combined using Bayesian method.
-
Investigated numerical performance of various surrogate models for 1-D predictions and found Ordinary Kriging was most reliable to avoid large error for deterministic functions.
-
Proposed a heuristic approach to determine the parameter of ridge regression for extrapolating data with noisy pattern.
Assessing usefulness of behavioral modeling & emulation for extreme-scale systems
-
Designed experiments for multi-level validation of different models to emulate execution time of a benchmark computational fluid dynamics (CFD) analysis run on Vulcan supercomputer
-
Proposed two metrics to assess usefulness of computational model for design optimization. The metrics quantifies two preferred features: (1) Trend preservation to accurately indicate relative magnitude between points; (2) Applicability of model correction against experiments.
-
Worked interactively with Computer Science (model A&B) and Exascale team (model C&D)