A new platform for laser cutters is capable of detecting and identifying thirty commonly used materials. This simplifies distinguishing between materials that are very similar.
The platform is called SensiCut and was developed by Mustafa Doga Dogan, Steven Vidal Acevedo Colon, Varnika Sinha, Kaan Akşit and Stefanie Mueller. They are all with MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL). Laser cutters enable the cutting of all kinds of materials, including metals, wood, paper and plastic. However, mishandling these materials can have unwanted consequences. For instance, the material may take an incorrect shape or give off odours. Harmful chemicals may also be released.
Sometimes difficult to distinguish
It is therefore important that users identify materials correctly. However, this is challenging in practice. For instance, some materials look very similar and are not easily distinguishable with the naked eye. Systems that can detect such materials have been available for some time. Here, the systems often use a camera-based approach. MIT researchers point out that this approach has its limitations. And can, among other things, lead to misidentification of materials.
Another commonly used method is identification using QR codes. This involves applying QR codes to individual sheets of material. However, if this code disappears from the material - e.g. by cutting the material - it can no longer be used for this purpose. In addition, if an incorrect QR code is affixed to a material, the system misidentifies the material. Such a system is therefore prone to human error.
Speckle sensing
SensiCut takes a different approach. The platform combines deep learning and an optical method called speckle sensing. SensiCut deploys a laser to map the microstructure of a material. In doing so, the platform looks, among other things, at the reflection of its laser beams on the material. Based on this, the material is identified.
"By supplementing standard laser cutters with lensless image sensors, we can easily identify visually similar materials that are common in workshops and reduce waste," says PhD student Mustafa Doga Dogan. "We do this by using the surface structure at the micron level. This is a unique characteristic, even if it (ed: the material) is similar to another type. Without it, you would probably have to make an educated guess in a large database to the correct material name."
Neural network
Among other things, the platform uses a neural network. Such a network is composed of processors that act like artificial neurons. They form several separate layers, each of which analyses the data differently and passes on the results to the next layer of neurons. Using over 38,000 images, this network has been trained to recognise 30 material types. This allows SensiCut to recognise acrylic, foamboard and styrene, among others. In addition, the platform can advise on settings of the laser cutter to cut the material adequately.
The system is based on low-cost components that are commonly available. As an example, the researchers cite a Raspberry Pi Zero single-board computer. These components are housed in a 3D printed lightweight enclosure and integrated with a laser cutter. The researchers deploy loose hardware, as most laser cutters are closed-source. They point out that adding this hardware to laser cutters is usually not necessary for manufacturers. They can use the computing power already present in laser cutters to perform the necessary data processing.
SensiCut also provides support for cutting and engraving objects composed of multiple materials. Traditionally, users have to manually split such a design into multiple files by material type. SensiCut automates this process and can automatically split the design.
Possibly also suitable for other devices
The team's work is focused on laser cutters. However, the researchers expect that SensiCut can eventually be integrated with other devices, such as 3D printers. They also want to expand SensiCut's capabilities. Among other things, the system should be able to detect material thickness.
Author: Wouter Hoeffnagel
