At some point in the future, some form of artificial intelligent entity will provide us with the automatic design of (meta)materials and structures with as much human intervention as we allow. We could then fully delegate the laborious part of automated design to a computer and focus all our attention to the uniquely human part. I call this autonomous virtual prototyping, and my research aims at achieving just that.

Even with the remarkable evolution in computational mechanics, current technology cannot be used to realize my vision due to the following main three roadblocks: i) modeling; ii) optimization; and iii) computational scalability. These issues limit the applicability of TO for virtual prototyping (see figure below). The trade-off between problem size and problem complexity is clear: While TO has been widely used to optimize simple problems, increasing the problem complexity reduces the size of problems that can effectively be confronted.

Problem size vs. problem complexity schematic shows that large complex problems could be confronted by harnessing the power behind AI, advanced analysis, and optimization.

Current technology therefore cannot be used to fully automate computational design. A breakthrough in virtual prototyping can only be attained by looking at computational design from a completely different angle. My approach combines advanced FEA—for fully automating analysis and optimization—with artificial intelligence (AI) algorithms. In particular, I explore the use of enriched FEA in topology optimization, combined to surrogate models based on machine learning (ML).