Visual Characteristics Of The Melt Pool In The High-nitrogen Steel CMT Additive Process And Forming Quality Evaluation Method

The molten pool of metal additive process contains rich quality information.Through the quality evaluation of the molten pool information detected,the real-time regulation of additive process can be realized,so as to ensure the quality of additive components and reduce the cost.Aiming at the demand of arc additive quality control of high nitrogen steel components,the research on robot cold metal transition technology(CMT)arc additive process,online real-time visual sensing of molten pool,feature extraction of molten pool image formed by defects in additive process,additive forming quality modeling and other technical research were carried out to provide technical methods and experimental data for real-time monitoring and quality evaluation of CMT arc additive for high nitrogen steel components.Robotic CMT additive process test for high nitrogen steel was carried out,and the most stable and best formed process parameters for the additive process were given.The characteristic information of CMT additive process of high nitrogen steel was collected by means of high-speed camera.It was found that the arc combustion cycle was 13.4ms,and the additive process included three droplet transition modes:drop transition,short-circuit transition and explosion transition,and the peak and base current/voltage ratio were 40.9%and 36.8%,respectively.The effects of wire feeding speed,additive speed and other process parameters on the forming size and surface defects of additive weld were explored.Passive vision and near-infrared filter were used to collect molten pool images under different additive dimensions.Taking the molten pool image in the mode of single-layer and multi-channel additive as the research object,the shape description parameters of molten pool image were proposed.A molten pool image processing algorithm based on threshold segmentation and morphology and a molten pool image processing algorithm based on the U-net semantic segmentation model were designed to obtain the contour information of the molten pool image,where the latter is more robust to noise such as metal vapor and wire occlusion.The experiment of high nitrogen steel robot CMT additive surface forming was carried out,and the influence of channel spacing on additive surface forming quality was explored.The geometric characteristic parameters of molten pool front,such as weld length,weld width,weld pool area,weld pool perimeter and weld pool symmetry,were proposed.By extracting the geometric characteristic parameters of weld pool image under different channel spacing,the variation law of the above parameters with channel spacing was summarized.The surface flatness s_qwas defined,and the surface flatness sq of the additive component under each channel spacing was calculated by three-dimensional scanning.Based on the s_qvalue and channel spacing,the additive surface flatness was divided into three categories:surface flatness,surface roughness and track spacing is too large,surface roughness and track spacing is too small.A molten pool visual image-driven prediction model for CMT additive surface flatness category of high nitrogen steel was developed,which takes molten pool image as input and additive surface flatness category as output,and was optimized to achieve 95%prediction accuracy on the test set.The influence of process parameters on weld porosity in CMT additive of high nitrogen steel was studied by means of X-ray flaw detection and pixel statistics,and the process parameters with high porosity were selected to collect the image of molten pool with porosity defects.In view of the characteristics that the visual characteristics of pore defects are not obvious and difficult to quantify,it was proposed that the region reflecting pore characteristics in the molten pool image is the semi solidified region,and the method to determine the molten pool image of pore defects was given.The pore prediction model of high nitrogen steel CMT additive process driven by molten pool visual image was established,by comparing the effects of different pre training models and full connection layer on the prediction accuracy,the model was optimized,the accuracy of the final model on the test set was 82.8%.A modified algorithm was proposed for the prediction results,which can further effectively eliminate the prediction deviation and realize the accurate prediction of pore defects.