Cutting Risk

A computer vision-based approach to material classification makes laser cutting safer and more efficient.

Nick Bild
3 years agoMachine Learning & AI
SensiCut (📷: M. Dogan et al.)

Laser cutters are very high-precision tools that can be used in cutting most materials. However, as you might expect, vaporizing things with high-powered lasers can be hazardous. Now that laser cutters have moved beyond industrial manufacturing applications, and can commonly be found in schools, small businesses, and on hobbyist workbenches, it is more important than ever to implement some appropriate safeguards.

Researchers at MIT CSAIL have put forward one approach to improve laser cutter safety with their SensiCut method. Laser cutter settings — such as power and speed — must be configured correctly for the type of material one is attempting to cut. Moreover, laser cutters should not be used at all for certain types of materials, because doing so will release toxic fumes. The problem is that many materials are virtually indistinguishable visually, and so it can be nearly impossible to tell exactly what those materials are in the scrap bin.

The SensiCut approach involves attaching a camera and laser pointer to an existing laser cutter. The laser pointer is shined on the material before cutting starts, and the speckle pattern it creates is captured by the camera. You may wonder why the laser pointer already onboard the cutter was not used — this is because the laser pointer onboard most cutters is red, and since most commercial cameras have Bayer filters that make them less sensitive to red, the team needed a green laser pointer to achieve a more clear speckle pattern. For th image sensor, an 8 megapixel Raspberry Pi camera module was chosen. The camera lens was removed to enhance its ability to capture speckle patterns.

A Raspberry Pi computer is included in the hardware add-on to capture images and transfer them to an external computer for further processing. This external computer runs a convolutional neural network, developed in PyTorch, that classifies the materials detected in the images. Appropriate settings for the cutter’s laser power, speed, and pulses per inch are retrieved from a database and are used to program the machine.

SensiCut was found to perform quite well, with an average classification accuracy of better than 98% across thirty different materials. Despite these excellent classification results, there is still work to be done on the device. Scratches on materials, for example, have a tendency to cause classification errors. The team is exploring capturing multiple images of the material to cross-check the results.

With a bit of refinement, SensiCut has the potential to offer a relatively simple, low-cost solution to accurately identifying materials. As laser cutters wind up being used in more situations, this may help to make them both safer and more efficient.

Nick Bild
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles