Courtesy of the researchers
MIT researcher have trained a machine-learning system
New materials that can be used in 3D printing are being developed constantly, but figuring how to print with them can be a complex, costly conundrum. Often, operators must use manual trial and error, possibly making thousands of prints to determine ideal parameters that consistently print the new material effectively.
MIT researchers have now used artificial intelligence to streamline the procedure. Scientists at the institution have developed a new machine-learning system that uses computer vision to watch the manufacturing process and then correct errors in how it handles the material in real time.
The work avoids the prohibitively expensive process of printing thousands or millions of real objects to train the neural network. It could also enable engineers to more easily incorporate novel materials into their prints.
Incorporating novel materials into prints could help engineers develop objects with special electrical or chemical properties. They could also be helped to adjust the printing process on-the-fly if material or environmental conditions change unexpectedly.
“This project is really the first demonstration of building a manufacturing system that uses machine learning to learn a complex control policy,” says senior author Wojciech Matusik, Professor of Electrical Engineering and Computer Science at MIT.
Matusik, who leads the Computational Design and Fabrication Group (CDFG) within the Computer Science and Artificial Intelligence Group (CSAIL), continued: “If you have manufacturing machines that are more intelligent, they can adapt to the changing environment in the workplace in real time, to improve the yields or the accuracy of the system. You can squeeze more out of the machine.”
Developing a machine-learning system comes with plenty of challenges, the researchers needed to measure what was happening on the printer in real time. They developed machine-vision system using two cameras aimed at the nozzle of the 3D printer.
The system shines a light at material as it is deposited and, based on how much light passes through, calculates the materials thickness.
Mike Foshey, a mechanical engineer and project manager in the CDFG, said: “You can think of the vision system as a set of eyes watching the process in real time.”
The controller would process the images received from the vision system and based on any error it sees, would adjust the feed rate and direction of the printer. Training a neural network-based controller to understand this is data-intensive, requiring making millions of prints. So, a simulator was built instead.
A process known as reinforcement learning was used to train the controller. The model learned through a trial-and-error with a reward method, where the model selected printing parameters and after being shown the expected output, was rewarded when the parameters chose minimised the error between print and expected outcome.
In the real world, conditions typically change due to slight variations or noise in the printing process. The researchers created a numerical model that approximates noise form the 3D printer, which they used to add noise to the simulation to get more realistic results.
“The interesting thing we found was that, by implementing this noise model, we were unable to transfer the control policy that was purely trained in simulation onto hardware without training with any physical experimentation,” said Foshey. “We didn’t need to do any fine-tuning on the actual equipment afterwards.”
The researchers say that when the controller was tested it printed more accurately than any other control. The control policy also learns how materials spread after being deposited and adjust parameters accordingly.
The next steps are for the researchers to develop controllers for other manufacturing processes. They want to see how the approach can be modified for scenarios with multiple layers of material, or multiple materials being printed at once. Future iterations could also use AI to recognise and adjust for viscosity in real-time.
MIT has a long history with additive manufacturing, and has spawned multiple major 3D printing companies, such as Desktop Metal and VulcanForms.
