Straightness deviation detection during ultrashort-pulsed laser robot cutting using machine learning and deep learning

The authors report artificial intelligence-based methods for detecting cutting straightness deviations in the flexible ultrathin glass cutting process using an ultrashort-pulsed laser robotic system. In this system, an ultrashort-pulsed laser is integrated onto the link behind the elbow of a six-axis articulated industrial robot. The laser beam is guided by a mirrors and propagates along subsequent robot axes, enabling precision laser processing over a large and flexible working area. System vibration is a critical factor affecting process quality and consistency. To monitor this dynamic behavior of the laser robot system during the cutting process, a triaxial accelerometer is mounted on the final robot link. The acquired vibration signals are applied to train machine learning and deep learning models to detect cutting straightness deviations. Random Forest, eXtreme Gradient Boosting, Light Gradient Boosting Machine, and Linear Discriminant Analysis, a lightweight 1D-convolutional neural network (CNN) model, a 1D-CNN autoencoder, and a transformer model are evaluated. Comparing time-domain and frequency-domain datasets, the lightweight 1D-CNN model trained on the frequency-domain vibration dataset achieves the best performance in detecting both normal and abnormal vibrations during the cutting process, with an AUC (area under the curve) of 0.93 and an F1-score of 0.844, demonstrating its capability to detect cutting straightness deviations in glass cutting processes using the ultrashort-pulsed laser robot system.