YuMing Zhang
University of Kentucky, KY·Engineering
Welding joins the metal in cars, ships, pipelines and bridges, but good welds still depend on skilled people who constantly adjust to changes in the metal, the joint shape and the heat. Robots can weld, yet most simply replay a programmed path with fixed settings and fail when conditions shift.
- Fields
Engineering
Computer Science
Materials Science
- Topics
arc welding
robotic welding
weld pool
- Grants
- 2 known grants, through Sep 2029
What they work on
- Infer compact descriptions of molten weld-pool behavior from high-speed sensor data using dynamic variational autoencoders and deep Bayesian filtering, neural methods that extract and track hidden states amid noise.
- Model how those hidden weld states evolve in continuous time under varying process conditions with neural ordinary differential equations, networks that learn equations of motion from data.
- Build a hybrid controller that combines generative adversarial imitation learning, which copies expert welders from demonstrations, with iterative linear quadratic regulation for real-time accuracy, stability and disturbance rejection.
- Validate the sensing-and-control framework on robotic welding platforms and in virtual-reality human-in-the-loop studies across multiple arc welding processes, materials and joint geometries.
- Fuse optical, infrared, electrical, acoustic and inertial signals into a single physics-grounded latent state, using dynamic 3D Gaussian representations to estimate weld geometry and temperature despite arc glare and smoke plume occlusion.
- Develop context-conditioned world models that predict how weld states evolve under different actions, machines, materials and operator styles, supporting multi-step rollouts with uncertainty estimates.
- Create generative planning algorithms in which diffusion- or flow-based policies propose action trajectories, evaluate them through world-model rollouts, and enforce differentiable safety constraints.
- Measure generalization of the anticipatory system across multiple arc welding processes and robot platforms using benchmark datasets and held-out transfer tests.
More about the research
Two linked projects ask how a robot can watch the glowing pool of molten metal and respond the way an experienced welder does, and even predict what the weld will do before acting. Welding serves as a demanding testbed: the scene is hot, bright, partly hidden by smoke, and changes in fractions of a second.
Why it matters
Skilled welders are in short supply and much welding work is hot, fumy and hazardous. Robots that sense and anticipate weld behavior could take on variable jobs in shipbuilding, energy infrastructure, transportation and heavy equipment, raising throughput and weld consistency while cutting rework and material waste and keeping workers out of dangerous positions. The sensing, prediction and planning methods could also transfer to other robots that must act on partly hidden physical processes.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Rui Yu
University of Louisville, KY·Computer Science
Extract a labeled structural skeleton from single noisy images by fitting geometric primitives such as planes and cylinders to monocular depth estimates, using large vision-language models to assign meaning.
Yusu Wang
University of California-San Diego, CA·Computer Science
Explain why trained flow-based and diffusion generative models generalize better than the theoretically optimal solution predicts, using geometry, ODE analysis, manifold learning and deep learning theory, to enable controlled generation of novel samples.
Steven Brunton
University of Washington, WA·Engineering
Develop physics-aware machine learning methods that build physical laws and stability constraints into the models, producing interpretable models that work reliably for aerospace design and control.
The school
University of Kentucky: 34 labs on PI Pages, $528M a year on research, 319 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.