Civil and environmental engineers Dat Ha and Josephine Carstensen call the approach “human-informed topology optimization.” In their field, AI-based generative design systems have grown in popularity as a method of producing complex architecture in a short amount of time. While useful, these systems aren’t perfect: They often overcompensate when it comes to load-bearing requirements, or they fail to provide the greatest possible strength in proportion to material weight. Meanwhile, having a human engineer design similar structures is time-consuming (and, when it comes to labor, quite expensive).
Human-informed topology optimization is the MIT engineers’ way of combining AI’s efficiency with human expertise. Rather than allowing an automated design system to generate a structure from start to finish, Ha and Carstensen periodically paused the system to evaluate its design and make adjustments. By introducing their expertise at various phases throughout the design process, Ha and Carstensen received better results than if they’d allowed the AI to generate a structure independently.
(Credit: Sven Mieke/Unsplash)
The engineers tested their approach with load-bearing beams, which would normally be used in a building or a bridge. They started by feeding the automated design system a handful of necessary criteria, like the beam’s required length, support points, and load-bearing capabilities. Then they initiated the system’s design process. When they noticed faults, they paused the system and revised those areas or instructed the system to conduct its own revisions.
Before you start questioning the structural integrity of that new building that just went up in your city, don’t worry—engineers tend to use “fully rigorous yet significantly slower design [algorithms] that consider the underlying physics” for structures that end up making it into the real world. But quicker, less-rounded systems like those supplemented by Ha and Carstensen still have their uses. In a release for MIT, the engineers note that disaster zones, war zones, and businesses designing small equipment tend to require simpler automated design systems due to their lower cost and lighter training requirements. Intercepting those systems’ design processes, however, could produce stronger results without requiring nearly as much labor as a fully human design process.
Now Read:
- Biocomputers Using Human Brain Cells Could Overpower AI
- Sony Unleashes Experimental AI Driver in Gran Turismo 7
- Microsoft Puts Its Bing AI Chatbot in the Windows 11 Taskbar
sourse ExtremeTechExtremeTech https://ift.tt/GpghVk8
ليست هناك تعليقات:
إرسال تعليق