Real-time control of laser materials processing using deep learning

Abstract

The plasma that is generated during laser materials processing can prevent the direct observation of the target. However, the appearance of the generated plasma is correlated with the properties of the material being ablated. Here, we show that deep learning can enable the identification of the material in real-time directly from processing camera images of the plasma, and hence can be used to automatically prevent machining beyond material boundaries. This work could have applications across laser materials processing in cases where the laser induced plasma restricts direct observation of the sample.