Additive manufacturing; composition monitoring; optical emission spectroscopy (OES); support vector regression (SVR);INDUCED BREAKDOWN SPECTROSCOPY; METAL-DEPOSITION; MAGNESIUM ALLOY; ARC PLASMA; ELEMENTS; DEFECTS; NETWORK; SAMPLES; SYSTEM; LIBS

Laser additive manufacturing has gained widespread adoption in recent years. However, process diagnosis and process control lag behind the progresses of other key technologies, which make the product quality control a challenging problem. This work proposes an operating parameter conditioned support vector regression (SVR) method that uses processing parameter conditioned kernel function to achieve a processing parameter independent in-situ composition prediction. Two different features of laser-induced plasma, spectral line-intensity-ratio, and both spectral line-intensity-ratio and spectral integrated intensity were used to train the SVR. Composition measurements using a calibration curve method, partial least square regression, and artificial neural networks are also performed for comparison. The results show that the SVR with both spectral line-intensity-ratio and spectral integrated intensity as inputs has the best performance due to linearly separable point clusters in the high-dimensional space. Laser power independent composition prediction is achieved and real-time composition predictions are validated. It is proved that the operating parameter conditioned SVR provides a more accurate, a more universal, and an operating parameter independent prediction. Moreover, operating parameter conditioned SVR provides a potential data-driven-based approach for real-time composition monitoring of the laser additive manufacturing process.