In High-Speed Laser Blanking for automotive production, efficiency is evident compared to traditional methods like stamping, as tool changes and prototype development are eliminated. Continuous processes in laser beam cutting require reliable detection of process boundaries to prevent downtimes and enhance productivity. AI-based real-time monitoring can optimize speed safety margins. Key technologies such as Minimal Invasive Laser Power Modulation (MILM) facilitate precise process monitoring and cut failure detection. A Convolutional Neural Network (CNN) achieves a classification accuracy of 97.9% and ensures real-time capability in operation. In application, the error rate of this classification result can be further reduced by implementing a hysteresis function to increase classification reliability, avoiding unnecessary downtime and opening up avenues for further process optimisation.

Precision in manufacturing has become essential in today’s competitive market, necessitating the minimization of all sources of inaccuracies. In the laser cutting process, the inherently thermal nature introduces significant heat to the workpiece, causing thermal expansion and dimensional errors in the absence of compensating strategies. These issues are particularly pronounced in aluminum tube workpieces due to their high thermal conductivity, large expansion coefficient, and extended axial dimensions of the raw material, which amplify thermal expansion effects. This study addresses these challenges by developing a real-time-capable predictive dynamic model. The model correlates commanded laser power with average heat-induced temperature increase, enabling precise, workpiece-specific thermal expansion estimation while maintaining computational efficiency. Calibrated and validated on an industrial laser tube machine, the proposed strategy achieves an average error reduction up to 75%, significantly improving dimensional accuracy and offering a robust solution for high-precision laser-based manufacturing.

To meet stringent requirements in terms of productivity and efficiency, the industry is shifting towards machine tools with sensors and auto-tuning capabilities enabled by Artificial Intelligence (AI) algorithms. Accordingly, in laser cutting the coaxial monitoring of the molten pool provides relevant information that can be interpreted by means of Machine Learning (ML) approaches to enable real-time velocity-based feedback control. In the present research, a holistic control architecture was thus developed and validated on an industrial case-study to demonstrate its applicability. Targeting iso-quality conditions, experiments on sample geometries on 5 mm thick stainless steel allowed to showcase an increase in productivity whilst avoiding critical defects such as cut dominated by plasma formation or loss of cut. The results obtained may be extended to a wide range of materials and sheet thicknesses demonstrating the generalized applicability of the technological framework.

Laser piercing is the initial phase of laser cutting, where a focused laser beam creates an entry hole in the material for subsequent cutting. This process is particularly challenging in thick stainless steel plates, as it requires precise control of energy input: excessive energy causes melt accumulation and process failure, while insufficient energy prolongs piercing time, reducing productivity. Prior studies demonstrated that time-varying pulse sequences can significantly decrease piercing duration, but identifying the optimal sequence within a few trials remains unresolved. This work presents a Bayesian optimization approach to minimize piercing time, demonstrated on 30 mm thick stainless steel plates with a 12 kW industrial solid state laser. To further refine the process, we integrate physically motivated strategies into the optimization framework. Our findings show that incorporating this additional knowledge can reduce the number of required optimization iterations, ultimately enhancing productivity in laser cutting operations by shortening process times.