Simulation Software and 3D Microstructure Skeletons Could Automate Materials Design

The innovative researchers at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) have been focused on topology optimization for 3D printing for quite some time, which makes the 3D design process easier. Designers have precise control over a 3D printed object’s microstructure, as 3D printers usually have a 600 dot per inch resolution. However, it would take too long to evaluate the physical effects of all possible combinations for an object, which is where topology optimization comes in. Last year, a CSAIL team designed a system that catalogs the physical properties of a large number of tiny clusters of voxels, which make up larger 3D printable objects.

Now, the same team members – Associate Professor of Electrical Engineering and Computer Science Wojciech Matusik, electrical engineering and computer science graduate student Desai Chen, and postdoc students Mélina Skouras and Bo Zhu in Matusik’s group – have developed a new approach for automating materials design, using their previous research as a jumping off point.

Materials scientists have long been inspired by the natural world, reverse-engineering traits like the toughness of conch shells to determine the microstructure of the material before working to make it with human-made materials. The CSAIL team’s new approach lists a material’s desired properties, and a 3D structure is then generated on a computer system, which places the microstructure design on what CSAIL refers to as “much more secure empirical footing.”

The team published a paper on their results, titled “Computational Discovery of Extremal Microstructure Families,” in the journal Science Advances; Chen is the first author.

The abstract reads, “Modern fabrication techniques, such as additive manufacturing, can be used to create materials with complex custom internal structures. These engineered materials exhibit a much broader range of bulk properties than their base materials and are typically referred to as metamaterials or microstructures. Although metamaterials with extraordinary properties have many applications, designing them is very difficult and is generally done by hand. We propose a computational approach to discover families of microstructures with extremal macroscale properties automatically. Using efficient simulation and sampling techniques, we compute the space of mechanical properties covered by physically realizable microstructures. Our system then clusters microstructures with common topologies into families. Parameterized templates are eventually extracted from families to generate new microstructure designs. We demonstrate these capabilities on the computational design of mechanical metamaterials and present five auxetic microstructure families with extremal elasticmaterial properties.”

New software identified five different families of microstructures, each defined by a shared “skeleton” that optimally traded off three mechanical properties.

Designers using this new system can numerically specify the properties they want for their materials, and it will automatically generate a 3D microstructure that fits. The team describes producing microstructures with optimal trade-offs between three mechanical properties in their paper, but Matusik says the approach is adaptable.

“We did it for relatively simple mechanical properties, but you can apply it to more complex mechanical properties, or you could apply it to combinations of thermal, mechanical, optical, and electromagnetic properties. Basically, this is a completely automated process for discovering optimal structure families for metamaterials,” Matusik explained.

In last year’s research, the team generated computer models of microstructures and scored them with simulation software, according to measurements of a few mechanical properties. They built a cloud of points (defined by each score), and each one corresponded to a microstructure. Then, they contained the dense cloud within a computed bounding surface, and nearby points were representative of good trade-offs between the properties.

At this point, the team’s new research picks up, and they evaluated the geometric similarities between the microstructures, using standard measures, that correspond to the boundary points. Then, their software clusters together microstructures with similar geometries based on these measures.

A rudimentary 3D shape shared by the microstructures, called a skeleton, is extracted for every cluster from the software. By making small adjustments to the skeleton, as well as building boxes around its segments, the software then attempts to reproduce the microstructures: basically conceiving of a mathematical formula to reconstruct each one in a cluster.

Machine learning techniques are used to discover correlations between the variables’ values and the microstructures’ measured properties, which helps the system translate between microstructures and their properties.

According to Matusik, nearly every step in the process, including clustering, skeleton extraction, and the formula derivation, is automated. But the system could also be paired with current materials design approaches, and either way could be the beginning of design possibility exploration.

Matusik said, “You can throw this into the bucket for your sampler. So we guarantee that we are at least as good as anything else that has been done before.”

However, the one part of the analysis that’s not automated is identification of the physical mechanisms that decide the properties of the microstructures: once skeletons of different microstructure families were available, the researchers were able to find out how they would respond to physical forces. But Chen said even this aspect could eventually be automated, as the simulation software could determine which structural elements will deform the most under physical pressure.

Discuss this and other 3D printing topics at 3DPrintBoard.com or share your thoughts below.