MIT researchers demonstrate stronger structural designs with human feedback

MIT researchers demonstrated a system that lets engineers modify structures during algorithmic design, improving manufacturability and increasing linear buckling load by 39 percent. Their study in Communications Engineering found that adding human feedback increased total design time by only 15 seconds compared with all-algorithmic approaches. The system uses machine learning to identify regions where a designer would likely want changes. It works with topology optimization, which iteratively distributes material within a design to meet specified goals. The model learned from synthetic feedback on thousands of designs. Users can currently make structures only thicker or thinner within the identified regions.