Research On Visual Inspection And Control Method Of Molten Pool For Arc Additive Manufacturing

Wire and arc additive manufacturing is based on the principle of layer-by-layer deposition,the metal wire is melted by arc heat,and under the control of the software program,according to the three-dimensional digital model,the advanced digital manufacturing technology of the metal components is gradually formed from the linesurface-body.Compared with traditional castings and forgings,it does not require molds,has high efficiency,low cost,and short production cycle.It has in-situ composite manufacturing and repair capabilities.The manufactured metal components have uniform composition and good mechanical properties,which can solve the problem of large metal components in the casting process.Problems such as component segregation,complex process,and high cost,but the arc additive manufacturing process is interfered by many factors,the size of the molten pool is difficult to maintain uniformity,the forming accuracy is low,and the existing molten pool detection methods rely on empirical parameters and the process is cumbersome,the accuracy rate is low,and the recognition time is long.Based on the above two problems,this paper detects the size of the molten pool in real time and forms a feedback control system.First,use the Res Net50 network to extract the characteristics of the molten pool,thereby generalizing the image characteristics of the molten pool,and get rid of the dependence on empirical parameters;then add the Point Rend neural network module to reduce the time required for molten pool detection;use the edge angle additional loss function instead the Softmax function makes the features extracted by the network more separable;the experimental results show that satisfactory detection results can be obtained with and without droplet occlusion.The detection width of the molten pool is compared with the measured width.The maximum absolute error is 0.36 mm,and the average value is 0.18 mm.The error is small,but the detection speed is slow,and only10 frames of images per second can be detected.Secondly,in order to improve the detection speed,a new molten pool shape detection algorithm is proposed,a first-level network is selected,the GIoU loss function is used instead of IoU to improve the accuracy of molten pool detection,and the convolutional layer and the BN layer are combined to improve the real-time performance of molten pool detection is verified by ablation experiments;finally,the detection width of the molten pool is compared with the measured width through visual analysis,the maximum absolute error is 0.39 mm,and the average value is 0.2mm.The error is small,within the acceptable range and the detection speed can reach 191.9frames per second.Third,multi-layer single channel test was performed to analyze the process parameters that affect the width of the molten pool.The width of the molten pool increases with the increase of the stacking current,and decreases with the increase of the welding speed and laser power.Based on the stacking current and the width of the molten pool has a good linear relationship and is easy to control,the accumulation current is selected as the single parameter to adjust the width of the molten pool;the current step response test is carried out,and the transfer function with the accumulation current as the input and the change in the width of the molten pool as the output is identified.A fuzzy PID control system is used,and the weld pool constant width,variable width,and variable welding speed control experiments show the necessity and effectiveness of the control system.Finally,the lightweight molten pool shape detection method based on GIoU and CONV?BN is used to detect the size of the molten pool when cladding the stern frame.When there is an error between the detection width of the molten pool and the set value,a fuzzy PID controller is used to correct the accumulation current during wire and arc additive manufacture was adjusted,and the results showed that the maximum size deviation of the stern shaft frame was 0.4mm,meeting the project index deviation within 1mm,with high forming accuracy,good surface quality,and no defects such as cracks and bubbles.