Researchers at MIT have discovered a means to create some of the best balances in modern manufacturing tools, such as 3D printers, by combining Artificial Intelligence (AI) and human manual design. AI can assist designers in creating structural materials in shapes that would be difficult or impossible to create with traditional methods.
However, such automated systems frequently fail to deliver designs fully optimised for their function, such as providing the greatest strength in proportion to weight or using the least amount of material to sustain a given load. In contrast, fully manual design is time-consuming and labour-intensive.
They used an automated design system but sometimes paused the process to allow human engineers to examine the work in progress and make changes or revisions before resuming the design process.
Introducing a few rounds resulted in outcomes that functioned better than those designed only by the automated method. The process was completed faster than the purely manual approach. The new generative design methods can use this flexibility to develop unique designs for components of a new building, car, or nearly any other item.
It’s a strategy for developing how to make things in much more complex ways than could be done in the past, Ha explains. Automated design systems have already become widely used in the automotive and aerospace industries, where reducing weight while maintaining structural strength is critical.
The researchers created a variety of structural load-bearing beams, such as those seen in a building or bridge, to demonstrate their approach in action. They identified an area in the design that could fail prematurely through iterations, so they chose that feature and forced the software to solve it. The computer system then altered the procedure to compensate, deleting the highlighted strut and strengthening specific other mounts, resulting in a better final design.
“You can take a lot of weight out of components, and weight drives everything in these two businesses,” he continued. In certain circumstances, such as internal components that are not visible, appearance is unimportant, while in others, aesthetics may be significant. The new method allows for the optimisation of designs for both visual and mechanical features, and human judgement is vital in such decisions.
The findings were published this week in the journal Structural and Multidisciplinary Optimisation in work co-authored by Dat Ha, an MIT PhD student, and Josephine Carstensen, an assistant professor of civil and environmental engineering.
According to Carstensen, the primary approach can be used for various scales and applications, including creating biomedical devices, nanoscale materials, and skyscraper structural support components. Automated design systems have already found numerous applications.
Not only can the technique be used to optimise a design for strength and weight, but it can also be used to optimise a design for any desired qualities. It can, for example, be utilised to reduce stresses in the material by softening corners or to minimise fracture or buckling.
“We’re not seeking to replace the seven-hour solution,” Carstensen explains. If you have all of the time and resources in the world, surely you can run these and get the optimal solution.” However, “then this kind of solution that addressed directly to your needs would prevail” in many instances, such as creating new components for equipment in a war zone or disaster-relief area with little computational power available.
Similarly, a simplified approach may be just what the doctor ordered for smaller organisations manufacturing equipment in essentially “mom and pop” shops. According to Carstensen, the new method they devised is simple and efficient to run on smaller systems and requires significantly less training to yield valuable results. Moreover, a rudimentary two-dimensional version of the software, sufficient for designing primary beams and structural sections, is now freely available online while the team works on a full 3D version.