Researchers at the Massachusetts Institute of Technology (MIT) have developed an AI-assisted design system that allows engineers to influence structural optimization while it is underway. The approach combines machine learning, topology optimization, and human feedback to produce structures that are easier to manufacture and better suited to practical engineering requirements, tells MIT News.
Traditional topology optimization uses algorithms to determine efficient material distribution within a structure. Although effective in aerospace, automotive, and other applications, the process can be computationally expensive and produce designs that are difficult to manufacture. Engineers typically review completed results, often requiring additional optimization cycles to address undesirable features.
MIT’s system addresses these limitations by introducing human judgment during optimization. A machine-learning model identifies regions where designers are most likely to recommend modifications. Engineers can then adjust those areas while the algorithm continues refining the structure.
The researchers trained the model using synthetic feedback on thousands of designs, incorporating considerations such as aesthetics, structural performance, and manufacturability. This approach allows engineers to contribute their expertise without significantly slowing the optimization process.
In demonstrations, the human-in-the-loop system improved manufacturability and linear buckling load by 39%, adding only approximately 15 seconds to the overall design time compared with fully algorithmic methods.
Currently, users can make selected structural regions thicker or thinner. The researchers plan to introduce additional modification capabilities in future versions.
The technology could be particularly valuable for buildings and bridges, where construction feasibility, structural reliability, and professional judgment influence design decisions. It may also encourage wider adoption of topology optimization among engineers who remain skeptical of automated design tools.
Rather than replacing engineers, the system positions AI as a collaborative design assistant. By incorporating human expertise throughout optimization, MIT aims to make algorithmically generated structures more practical, trustworthy, and accessible across engineering disciplines.
