Research Seminar on AI: Investigation of Reinforcement Learning for Shape Optimization of Flow Channels in Profile Extrusion Dies

 

Tuesday, October 18, 2022, 4:00pm

Profile extrusion is a continuous production process for manufacturing plastic profiles from molten polymer. Especially interesting is the design of the die, through which the melt is pressed to impart the desired shape. However, due to an inhomogeneous velocity distribution at the die exit or residual stresses inside the extrudate, the final shape of the manufactured part often deviates from the desired one. To avoid these deviations, Daniel Wolff and his team want to optimize the shape of the die computationally, which has already been investigated in the literature using classical optimization approaches.
In their work, Daniel Wolff and his team investigate the feasibility of Reinforcement Learning (RL) as a learning-based algorithm for shape optimization. RL is based on trial-and-error interactions of an agent with an environment. For each action, the agent is rewarded and informed about the subsequent state of the environment. While not necessarily superior to classical, e.g., gradient-based or evolutionary, optimization algorithms for one single problem, RL techniques are expected to perform especially well when similar optimization tasks are repeated, since the agent learns a more general strategy for generating optimal shapes instead of concentrating on just one single problem.
Daniel Wolff and his team will introduce a 2D test case, in which an RL agent can directly modify the geometry of a die flow channel's computational mesh through a spline-based deformation method known as Free Form Deformation (FFD) by manipulating the control point coordinates of the transformation spline. Every action of the agent requires the computation of the environment's new state by performing a high-fidelity Finite Element Method (FEM) simulation in the modified geometry. Based on this test case, Daniel Wolff and his team address different research questions with the goal of better understanding the capabilities of this new approach. They will compare different optimization approaches and learning algorithms as well as investigate possibilities to speed up the computationally expensive training. The presented findings pave the way for the application of this method to more complex, industrial geometries.

Daniel Wolff did his Bachelor's and Master's degrees in “Computational Engineering Science” at RWTH Aachen University. Afterward, he started his Ph.D. with the Chair for Computational Analysis of Technical Systems and the Helmholtz School for Data Science in Life, Earth & Energy (HDS-LEE), working on applying methods from the field of scientific machine learning to simulations in the field of profile extrusion and bioreactors.