Final project for my graduation envolving application of a LSTM neural network coupled with MPC in the optimization of a single bead geometry welded through WAAM. The MPC communicates with the outside world via ROS nodes.
This project has 3 components: \
-
Process model (PM): The model used to identify the process dynamics and modelling it, used in the MPC as the predictive model of the system.
-
Gradient model (GM): The model used to estimate the gradients of PM.
-
MPC: The controller itself, with the optimizer, the predictive model (PM) and the gradient model (GM).
The workflow in divided in 3 main parts: data generation, models training and MPC deployment in ROS.