Russian researchers have used machine learning to make metal 3D printing more efficient.
3D printers require fine tuning of positioning and control algorithms using mathematical models to reach optimal performance. This is a lengthy and arduous process and it could take weeks to set printing parameters. Even then, the possibility of printing error is always present.
To overcome such problems scientists at the Laboratory of Lightweight Materials and Structures of Peter the Great St. Petersburg Polytechnic University (SPbPU) have developed a neural network for a metal 3D printer.
Brainy neural networks
Whether in the form of collaborative robots, 3D printers using AI, software equipped with thinking capabilities, or enabling smart factories, moving towards industry 4.0, engineers and scientists have brought Artificial Intelligence (AI) to the manufacturing industry.
Going further, scientists are also exploring neural networks to make AI more capable in 3D printing.
A neural network is a framework for deep machine learning inspired by the connectionist architecture of the brain neurons. It is a mode of communication and feedback response which a machine uses to become capable of self-learning. Neural networks are different from AI algorithms, as they do not use task-specific rules.
Using a neural network, a computer can develop image recognition abilities, among others capabilities. For example, it can take a manual input of an image of ‘dog’ and ‘not-dog’. With this data, the machine could develop a criterion of difference between what is a dog and what is not a dog. And in future, use this as an image recognition criteria. Google image CAPTCHA is a neural network which works on this principle.
3D printing has also helped in the development of neural networks. As previously reported, scientists used 3D printing to build a neural network.
In the latest development, researchers at the SPbPU have developed neural network for metal 3D printing which will potentially make 3D printing faster and more efficient.
Self-learning 3D printer
Working on 3D printing project is not new for SPbPU. As previously reported, engineers at SPbPU created an electric engine with 3D printed components.
Now, the SPbPU team has a developed neural network that learns from previous manually entered data to make 3D printing faster, eliminating the need for mathematical modeling for different structures.
Furthermore, the neural network also makes an adjustment during the print to detect and amend flaws, a problem also addressed by Lawrence Livermore National Laboratory using neural networks.
The SPbPU neural network for 3D printing was developed in MATLAB, a numerical computing software, and a programming language.
Using the new neural network, SPbPU team developed printing modes to manufacture ship mastheads. SPbPU scientists are further testing the developed neural network. So far they have tested the quality of laser melting, the quality of manufactured parts, and the stability of the welding process.
Oleg Panchenko, Head of the Laboratory of Lightweight Materials and Structures SPbPU, elaborated “The next step is to create an online system based on the neural network with automatical input of data sets and output of parameters, thus such system will be learning continuously. We believe that the new system will improve the quality of the parts and increase the speed of parameters development for further manufacturing.”
A patent has been applied for.
Subscribe to our 3D printing newsletter for the latest news in 3D printing research. You can also visit our Facebook and Twitter pages.
If you are looking for a job in 3D printing then visit our 3D Printing Jobs.
Featured image shows an aerial view of the Peter the Great St. Petersburg Polytechnic University. Image via mun: planet
