In a cluttered, busy workshop, it's certainly possible to lose track of what substance a given sheet of material is composed of – making it risky to cut with a laser. The SensiCut system is designed to help, by identifying 30 different materials based on their surface qualities.
Currently being developed in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), the experimental setup can be added onto existing laser cutters. It consists of four main parts: a laser pointer, an image sensor, a Raspberry Pi Zero microprocessor and a battery. These are all contained within a 3D-printed housing.
After the material in question has been laid out on the laser cutting bed, the SensiCut laser is shone down onto it. The unique microstructure of the surface of the material causes the laser light to be reflected back up in a distinctive speckled pattern, which is picked up by the image sensor. Utilizing a deep neural network on a linked computer, the system is able to match that pattern to one of a known material.
A display on the computer screen subsequently tells the user what the material is, and indicates the ideal power and speed settings for the cutting laser. It also suggests the best uses for each material, plus it alerts operators if the material simply should not be cut with a laser – certain plastics may melt entirely, or give off particularly toxic fumes if laser-heated.
SensiCut can additionally laser-scan the whole surface of a flat object made of multiple materials, determining which areas are made of which substances. It can then guide the laser cutter as it engraves text or graphics onto that object, automatically adjusting the power and speed of the laser as it moves back and forth between the different materials.
The technology is currently 98 percent accurate at identifying materials such as different types of plastic, metal, wood and paper. By contrast, existing systems that simply utilize an optical camera to assess a material's visual characteristics are claimed to be much less accurate.
SensiCut is demonstrated in the video below.
Source: MIT
