We may be on a cusp of the transition to a world of complex, customizable products manufactured on demand by flexible robotic systems. But essential to this transition will be the new field of AI-based computational design for manufacturing, according to Wojciech Matusik, associate professor of electrical engineering and computer science at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT, where he leads the Computational Fabrication Group. Matusik will address the subject in a three-day MIT Professional Education course July 16 to July 18, 2018, on the MIT campus in Cambridge, MA.
The shift is being driven in part by the emergence of 3-D printers and whole-garment knitting machines, which open the door to rapid prototyping with a quick transition to full-scale manufacturing. However, said Matusik in a recent phone interview, current design strategies cannot keep up with the ever-more-capable manufacturing infrastructure. Consequently, it is necessary to automate the design processes to address larger design spaces across multiple domains—electronic, optical, mechanical, and thermal, for example.
“When every product is different, we cannot have a designer work on each one individually,” Matusik said. “We have to develop a much more automated method to convert specifications to digital files to send to the various fabrication devices.”
And that method will rely much more on AI techniques, he said, to work across multiple domains and make optimal tradeoffs among multiple competing design objectives. Simulation tools can provide examples from which machine learning can take place. Ultimately, he said, “You are replacing very complex simulation techniques with machine-learning methods to improve performance for a given design.”
Matusik cited a technique he calls generative design that allows engineers to specify functionality rather than materials and shape—whose selection will be up to the automated tools. That will complement other design strategies—hierarchical design and concurrent design—that his course will address.
Matusik said it would be beneficial to reuse existing tools to the extent possible, but they’ll be wrapped around a more sophisticated AI computational framework, and completely new tools will be needed as well. Providing an example of how new and old design tools might work together, Matusik coauthored a paper last year presenting InstantCAD, a tool that integrates with existing CAD software as a plugin and lets designers interactively edit, improve, and optimize CAD models.
“From more ergonomic desks to higher-performance cars, this is really about creating better products in less time,” said the paper’s lead author, Adriana Schulz, as quoted in MIT News. “We think this could be a real game changer for automakers and other companies that want to be able to test and improve complex designs in a matter of seconds to minutes, instead of hours to days.” Schulz, a Ph.D. student in MIT’s Department of Electrical Engineering and Computer Science, presented the paper at last summer’s Siggraph computer-graphics conference in Los Angeles.
“In a world where 3-D printing and industrial robotics are making manufacturing more accessible, we need systems that make the actual design process more accessible, too,” Schulz added. “With systems like this that make it easier to customize objects to meet your specific needs, we hope to be paving the way to a new age of personal manufacturing and DIY design.”
For Matusik’s course, experience with specific CAD tools is not necessary, but students will need to bring a Windows laptop or tablet for which they have administrator privileges.
Matusik said during our phone interview that computational design will be driven both from the top down and bottom up, with CEOs and managers setting direction but with individual scientists and engineers bringing the necessary education and knowledge to implement the methodology. The course is designed for research scientists, engineers, developers, designers, and project managers working in automotive, robotics, aerospace, defense, shipbuilding, biomedical, and textile industries as well as those working in areas such as prosthetics manufacturing, mechanical engineering, product design, and computer graphics. Matusik said students will come away from the course with examples of how to use existing tools for computational design, but he would encourage them to follow up with additional study on how to create new workflows within their organizations.
In addition to the Siggraph paper on InstantCAD, several other recent initiatives suggest how the future of computational design for manufacturing might work. For example, yet another Siggraph paper coauthored by Matuzik described how designers can gain control of a 3-D-printed microstructure’s physical properties such as density or strength.
“Conventionally, people design 3-D prints manually,” said Bo Zhu, a postdoc at CSAIL and first author on the paper, as quoted at MIT News. “But when you want to have some higher-level goal—for example, you want to design a chair with maximum stiffness or design some functional soft [robotic] gripper—then intuition or experience is maybe not enough. Topology optimization, which is the focus of our paper, incorporates the physics and simulation in the design loop. The problem for current topology optimization is that there is a gap between the hardware capabilities and the software. Our algorithm fills that gap.”
And in January, Matusik and coauthors elaborated on this work with a paper in Science Advances, describing metamaterials as engineered materials with complex custom internal structures that exhibit a broader range of bulk properties than their base materials. “Although metamaterials with extraordinary properties have many applications, designing them is very difficult and is generally done by hand,” they write. “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.”
Larry Hardesty at MIT News summarized this latest work, describing how materials scientists have traditionally tried to reverse-engineer natural materials like bones or shells and develop manmade materials that mimic their toughness or other desirable trait. The researchers, he wrote, “…have developed a new system that puts the design of microstructures on a much more secure empirical footing. With their system, designers numerically specify the properties they want their materials to have, and the system generates a microstructure that matches the specification.”
Finally, Matusik’s course this summer will conclude with a lab session on integrated design and optimization of custom drones. A 2016 paper by Matusik and coauthors provides a preview of what this session might involve, describing a system that lets users design, simulate, and build a drone with their choice of size, shape, structure, payload, flight time, and other factors—avoiding a one-size-fits-all approach to drone design, according to Matusik. Ph.D. student Tao Du commented on the tradeoffs. “For example, adding more rotors generally lets you carry more weight, but you also need to think about how to balance the drone to make sure it doesn’t tip,” he said, as quoted in MIT News. “Irregularly-shaped drones are very difficult to stabilize, which means that they require establishing very complex control parameters.”
Matusik’s course is one of a new wave of MIT Professional Education courses designed to reflect the knowledge needs of the workforce of the future. MIT has added a total of seven new classes to its 2018 Short Programs portfolio covering emerging fields and technologies such as artificial intelligence, machine learning, automation, and computational design.
For more information on Matusik’s course and to register, visit MIT Professional Education.
Here are the other five courses in the new wave:
- Advanced Machine Learning for Big Data and Text Processing provides an in-depth look at the tools, techniques, and algorithms driving modern and predictive analysis. Instructors are Regina Barzilay, Tommi Jaakkola, and Stefanie Jegeika.
- Advances in Food Innovation examines opportunities, challenges, and potential solutions in agriculture and food innovation. Instructor is Markus Buehler.
- Machine Learning for Big Data and Text Processing: Foundations explores core mathematical concepts and theories relevant to machine learning. Instructors: Instructors are Regina Barzilay, Tommi Jaakkola, and Stefanie Jegelka.
- Machine Learning for Healthcare examines emerging trends in machine learning methods for healthcare and how they will shape policy and personalized medicine in years to come. Instructor is David Sontag.
- Modeling and Optimization for Machine Learning reduces engineering and computational problems to their standard mathematical forms to determine which algorithms and software tools will best solve them. Instructors are Justin Solomon and Suvrit Sra.