Research On Quality Monitoring Of Selective Laser Melting Process Based On Machine Vision

Selective laser melting technology is an advanced manufacturing technology that has developed rapidly in recent years and has been widely used in many industrial sectors such as aerospace,toolmaking,and medical.However,the selective laser melting process is a complex process with multiple physical fields coupled to each other and highly dynamic.The disadvantages of poor process repeatability and insufficient stability limit the further industrial application of this technology.How to improve the quality of manufactured workpieces has always been in this field research focus.The manufacturing process of the selective laser melting technology includes from the powder spreading process,the melting process to the formation of molten layer.The process monitoring technology can detect the molding state and defects in time,eliminating the limitations of the above process problems on the development of the technology.In this paper,machine vision technology is applied to the process monitoring of laser melting in selected areas,and the method of identifying defects in the manufacturing process is studied.The research content is divided into the following parts:First of all,according to the structure and working principle of the selective laser melting equipment,the common types of defects in the laser selective melting process are analyzed.Then,according to the actual needs,suitable hardware such as industrial cameras,lenses,and light sources were screened to build a machine vision monitoring system,and a paraxial shooting method was adopted for the equipment working area.Calibration and angle of view correction for industrial cameras.Secondly,the pre-processing and image segmentation methods of the powder bed image are researche.Using machine vision monitoring system to collect laser melting powder bed image.Due to the lighting problem,the shooting angle of the collection system,and the internal environmental impact of the actual equipment,factors that are not conducive to image processing cannot be avoided.Therefore,the research work mainly uses corresponding image processing algorithms to deal with defects from the powder bed image.lt mainly includes the uneven illumination processing of the image based on the two-dimensional gamma function,the local threshold segmentation algorithm based on the variance and gradient,the morphology and connected domain noise processing algorithm,and the Hough transform algorithm to extract linear defects.Finally,through pattern recognition,two methods for identifying defects in the powder bed image for the selective laser melting technique were studied.On the one hand,the research used feature extraction based on features such as morphology,area size,gray standard deviation,etc.,combined with C4.5 algorithm to construct the defect identification of the decision tree,and the average accuracy of the identified defects reached 91.05%.On the other hand,on the basis of the AlexNet convolutional neural network,the input and classification layers of the powder bed image suitable for this paper were improved,and the recognition of defect types in this paper was realized in combination with transfer learning.The verification shows that the recognition rate of the image defect of the improved AlexNet network in the powder bed can reach 95.92%for the linear defect of the collision fringe,and the recognition rate of the fragment defect is greatly improved by 83.04%,and the overall recognition efficiency is better than the original AlexNet neural network.