Offline planning optimization and formation prediction of laser directed energy deposition process

•The CTAR corner detection algorithm was used to the trajectory corners.•The Mask R-CNN was improved to molten pool target segmentation.•The Stacking ensemble learning model was used to predict the cladding layer geometry. Laser directed energy deposition process involves many complex physical and chemical changes, and there is a strong coupling relationship between various process parameters, which makes the final forming quality difficult to control. In this paper, the optimization strategy of forming quality was designed based on the application of laser-directed energy deposition of 316L stainless steel. The CTAR corner detection algorithm was used to precisely locate the position of the trajectory corners and adjust the laser power. Then the blade model was used to verify the optimization scheme. The experimental result showed that the bulging phenomenon of the blade corners can be eliminated. The deep learning model Mask R-CNN was improved to adapt to the scene of molten pool target segmentation, and the geometric feature parameters of the molten pool were obtained after the molten pool image was processed with this model. Finally, the Stacking ensemble learning model was used to establish a prediction model between the process parameters and the height of the cladding layer. This model had high reliability and guided the quality control of the laser-directed energy deposition process.