REAL-TIME MONITORING AND GAUSSIAN PROCESS-BASED ESTIMATION OF THE MELT POOL PROFILE IN DIRECT ENERGY DEPOSITION

A comprehensive understanding of the melt pool behavior during directed energy deposition (DED) has become essential in identifying process anomalies and controlling the process quality. Previous studies focused on predicting the melt pool characteristics by solely using the process parameters, in this study we use real-time melt pool images to predict the melt pool characteristics. A CMOS camera is used to capture coaxial images of the melt pool during the deposition of single-track prints to improve the prediction model. Multiple regression models are trained and compared to estimate the melt pool profile (width, depth, and height) as a function of process parameters (namely, the laser power, the powder feed rate, and the scanning speed) and features from the ellipse fitting of the real-time melt pool images. A novel image processing algorithm is proposed to extract the major axis, minor axis, and tilt. The sensitivity analysis demonstrated that combining process parameters and coaxial images can improve the prediction performance. The Gaussian Process regression showed the best performance among all the employed regression models.