GO-MELT: Thermal Modeling and Predictive Control

Thermal FEA, recurrent process models, and predictive control of laser power and beam mode in simulation.

I use GO-MELT thermal simulations to develop predictive controllers that adjust laser power and beam mode to regulate melt-pool geometry and photodiode temperature.

  • Configured locally refined meshes, laser heat sources, and thermal boundary conditions in the existing GO-MELT finite-element solver.
  • Developed a scheme to filter experimental data for the team’s simulation calibration and assessed mesh sensitivity.
  • Generated temperature and melt-pool histories to train recurrent prediction models using long short-term memory (LSTM) networks.
  • Developed model predictive control (MPC) and trained differentiable predictive control (DPC) for coordinated power and beam-mode commands.

In the simulation below, joint MPC and trained DPC followed the large width changes that the power-only controllers missed, while also regulating depth and temperature.

Trained controller comparison showing melt-pool depth, width, photodiode temperature, laser power, and beam mode over time
Simulation comparison of joint MPC, trained DPC, power-only MPC, and PID. The lower panels show the commanded laser power and beam mode.

View the full-size figure

Back to projects