Real-time continuous estimation of dross attachment in the laser cutting process based on process emission images

Laser cutting of metals has become the reference manufacturing technology in sheet metal working thanks to the flexibility and the increased productivity it offers when compared with other competitive technologies. Considering, in particular, the fusion-cutting mode, i.e., when nitrogen is used as an assisting gas, different aspects contribute to the process quality among which dross attachment plays the most important role. To cope with the related time-dependent deterioration of the process quality and to obtain an online adaptation of the process parameters for different working conditions, a closed-loop dross regulation system is needed. To realize it, a reliable, continuous, and accurate estimation of the dross is mandatory. This work focuses on this challenging problem, presenting and comparing different approaches to estimate the dross attachment based on the process emission collected by a coaxial camera. Specifically, a method which relies on the accurate analysis of the process emissions for determining an effective classification method is compared with a deep-learning approach based on convolutional neural networks. The obtained results, validated in real experimental conditions, confirm the possibility to accurately estimate the presence of significant dross attachment in real-time and open the way to the design of a closed-loop control algorithm for the real-time regulation of the dross attachment formation and consequently of the process quality.