18SC007 - Course on Structural Optimization

Instructor: Pedro V. Marcal

Multi-Objective Optimization for FEA by AI, via Q-learning and delayed reward (Watkins Theorem) combined with Design of Experiments. [1]

Watkins Theorem of Q-learning and delayed reward has shown exceptional results in model free AI problems such as the AlphaGo Go playing program.which beat the World Go Champion Lee Sedol. The theorem is even more powerful when it's used with problems where models exist, for instance FEA problems. The objective of the course is to explain Watkins theorem and demonstrate its application to the optimal design of structures for minimum volume. This problem is usually solved by a nonlinear programming process involving complicated coding and large scale computing. The approach using Q-learning requires a simple code written in Python to implement the theorem and the use of a FEA code. The course attendees will leave with an appreciation of how the method can be applied to a wide selection of FEA problems.