Laser oxidation cutting is a well-established technology in the metal-working industry for the cutting of high thickness mild steel. Cut quality is typically evaluated in terms of profile roughness, which normally decreases at high speeds. On the other hand as the cut velocity is increased loss-of-cut is more likely to occur hampering the cut completion. Hence, the cutting speed is typically set at a lower level yielding suboptimal quality and decreasing productivity. Although real-time estimation of the roughness profile through the analysis of process emission images has been explored in literature, it has not been implemented yet for quality optimization in real-time control applications. Within this framework, the availability of a speed-control architecture that regulates the cutting speed to minimize the roughness level while preventing loss-of-cut could enhance productivity while ensuring quality and process robustness.


This paper shows how such an architecture can be effectively designed and implemented. Specifically, we propose an online roughness estimator and regulation system coupled with a loss-of-cut supervisor. The control architecture was tested during the laser oxidation cutting of 8 mm thick mild steel on a machine endowed with a NIR coaxial monitoring system. Different machine learning algorithms were trained and tested to build effective prediction models for the selected quality indicators, yielding good fitting for both roughness estimation and loss of cut classification (with an R² value above 80% and an F1-score larger than 85% respectively). The real-life experiments witness the appropriateness of the proposed approach for industrial applications.