UPS researchers came up with an improved method for failure detection in 3D printers

UPS, viernes 21 agosto 2020
Improved method for failure detection in 3D printers
Improved method for failure detection in 3D printers

 

Professors at Universidad Politecnica Salesiana (UPS) have developed a new method to detect different types of failures in industrial 3D printers. The research, published in the Mechanical Systems and Signal Processing Journal - one of the most prestigious journals in mechanical engineering worldwide- will help industries that use 3D printing to avoid serious failures. 

"The problem with detecting failures is commonly approached by learning a binary classification model. An analogy to explain this would be when a child who is learning to recognize a dog from a cat, we would show the child images of the first animal indicating that it is a dog and then do the same with the second animal, exchanging between them. But what happens if we do not have an example of one of the two animals? This, in failure detection is equivalent to: what happens if we do not have examples of data with Failure to teach our model the difference? To make the things more complicated, if just like the senses (sight, hearing, etc.) of a human being, there is not just one perception mechanism but several ones? Those extra restrictions are not addressed by common learning methods", stated Cabréra

In their research, scientists devised a way to characterize signals from different variables such as position, velocity, acceleration and magnetic field through a set of Convolutional Neural Networks (bio-inspired by the visual cortex of some animals) created only with examples of the machine. in normal operation. They then created a fusion mechanism of these characteristics by means of a type of Vector Support Machines for learning from a single class of examples. "The performance evaluation of our proposal, grouping different types of sensors, showed considerable differences. However, in all cases our results were better than other reported approaches" commented Diego Cabrera, Mariela Cerrada and Vinicio Sánchez; professors who make up the university's Research and Development Group on Industrial Technologies, GIDTEC (for its acronym in Spanish).