New AI “copilot” helps engineers design better structures

A new AI "copilot" system developed by MIT researchers assists engineers in designing better structures, such as ultralight aircraft, high-bandwidth antennas, and F1 racing cars. However, traditional algorithmic design techniques demand significant computation and time, resulting in final designs that may be hard to manufacture or trust for structural integrity. In response, MIT researchers have created a system that functions like an AI copilot, integrating human feedback during the algorithmic design process.

The researchers' system, built on topology optimization techniques, presents designers with regions within the algorithm's output for review and modification. This collaboration reduces design time and increases manufacturability while maintaining structural performance. By training a machine-learning model with synthetic data reflecting probable human feedback, the researchers have developed an approach that brings together the strengths of algorithmic and human design methods.

In tests, the human-in-the-loop system improved manufacturability and linear buckling load by 39 percent without significantly increasing design time. The researchers aim to expand the approach's capabilities, allowing users to modify structures in various ways. By incorporating human input, the system also builds trust in the design process, ensuring that final structures are more likely to be accepted and constructed by industry professionals.