Using AI to correct additive manufacturing errors in real time

Researchers at the Massachusetts Institute of Technology (MIT) trained a machine learning model to monitor and adjust the 3D printing process to correct errors in real time.

New materials that can be used for 3D printing are constantly being developed, but figuring out how to print with them can be a complex and costly puzzle. Typically, operators must use manual trial and error, potentially running thousands of prints, to determine the ideal parameters to print new materials consistently and efficiently.

Researchers at MIT have used artificial intelligence to streamline the process. Scientists at the agency have developed a new machine learning system that uses computer vision to observe the manufacturing process and can correct errors in how materials are handled in real time.

They used simulations to teach a neural network how to adjust printing parameters to minimize errors, and then applied the controller to a real 3D printer. The new system can print objects more accurately than other existing 3D printing controllers.

This work avoids the expensive process of printing thousands or millions of real objects to train neural networks. It could make it easier for engineers to integrate new materials into their 3D printed products, which could help them develop products with special electrical or chemical properties. It also helps technicians make adjustments to the printing process when materials or environmental conditions change unexpectedly.

"This project is really the first demonstration of building a manufacturing system that uses machine learning to learn complex control strategies," said Wojciech Matusik, professor of electrical engineering and computer science at MIT who led the project. If you have smarter manufacturing equipment that can adapt in real time to changing circumstances in the workplace to increase output or system accuracy, you can get more value out of your machines."

Pick Parameters

Determining the ideal parameters for a digital manufacturing process can be one of the most expensive parts of the process, as it requires a lot of trial and error. Once the technician finds a combination that works well, these parameters only apply to a specific situation. They have little data on whether the materials behave in other environments, on different hardware, or on new batches of materials that exhibit different properties.

Using machine learning systems is also full of challenges. First, the researchers need to measure what is happening on the 3D printer in real time.

To do this, the researchers developed a machine vision system that uses two cameras aimed at the 3D printer’s nozzle. The system shines light onto the material as it is being deposited and calculates the thickness of the material based on the amount of light that passes through. "You can think of the visual system as a pair of eyes observing this process in real time," Foshey said.

The controller will then process the images it receives from the vision system and adjust the feed rate and orientation of the printer based on any errors it sees.

But training a neural network-based controller to understand this manufacturing process is data-intensive and requires millions of prints. So the researchers built a simulator.

Successful Simulation

To better train the controller, they used a process called reinforcement learning, in which The model learns through trial and error and is rewarded. The task of the model is to select printing parameters in order to create specific objects in the simulation environment. After showing the expected output, the model is rewarded when it selects parameters that minimize the error between its print and the expected result.

In this case, "error" means that the model either allocated too much material, placing it in areas that should remain open, or did not allocate enough material, leaving The open spots below should be filled. As the model performs more simulated prints, it updates its control strategy to maximize rewards, becoming increasingly accurate.

However, the real world is messier than the simulation. In practice, conditions often vary due to small changes or noise in the printing process. So the researchers created a numerical model that approximates the noise from a 3D printer. They used this model to add noise to their simulations, producing more realistic results.

"We found it interesting that by implementing this noise model we were able to convert controls trained purely in simulation The strategy is transferred to the hardware without any physical experiments for training," Foshey said, "and afterwards, we don't need to do any fine-tuning on the actual device."

When testing the controller It prints objects more accurately than any other control method previously evaluated. It performs particularly well in infill printing, which is printing the inside of an object. Some other controllers deposited so much material that the printed objects were raised, but the researchers' controller adjusted the printing path so that the objects remained level.

Their control strategy can even understand how the material spreads after deposition and adjust parameters accordingly.

auto-adjust

"We are also able to design control strategies that can dynamically control different types of materials. So if you have a manufacturing process on site and you want to change materials, you don't have to re-validate the manufacturing process. You can just load the new material, The controller will automatically adjust," Foshey said.

Now that they have demonstrated the effectiveness of this technique for 3D printing, the researchers hope to develop controllers for other manufacturing processes. They also want to see how the method can be modified to handle multiple layers of materials or printing multiple materials at the same time. Additionally, their method assumes each material has a fixed viscosity, but future iterations could use AI to identify and adjust viscosity in real time.

MIT has a long history in additive manufacturing and has spawned several major 3D printing companies, such as Desktop Metal and VulcanForms. This work was supported in part by the FWF Lise-Meitner Program, a European Research Council Starting Grant, and the National Science Foundation.

The above is the detailed content of Using AI to correct additive manufacturing errors in real time. For more information, please follow other related articles on the PHP Chinese website!