Online Monitoring Of Defects In Powder Spreading Process For Additive Manufacturing Based On Machine Vision

Additive manufacturing technology has been rapidly developed in recent years,however,as the representative powder-based additive manufacturing technology,the quality of selective laser sintering can be directly affected by powder spreading defects.A machine-vision-based defect detection method for coated sand SLS process is proposed to realize closed-loop control of manufacturing process and improve both quality and efficiency.The main research contents in this paper are as follows:(1)The object of this paper is to study four types of typical powder spreading defects: debris,recoater streaking,incomplete spreading and oversupplying.The images of abovementioned powder spreading defects has been collected by using an independently-built system,which includes AM system and machine-vision system.The setting method of lighting system for different defects is determined by comparing the degree of obviousness of the defect characteristics under different lighting directions.A image dataset of the SLS powder spreading process using coated ceramic sand is established.(2)In this paper,deep learning method is used to locate and classify the powder spreading defects,and after comparing various object detection algorithms,the YOLOv3 algorithm is chosen.The YOLOv3 model trained in this paper has an mAP of 95.83% and a Recall of 83.76% on the test set,and the average detection time per image is 16.6ms,which can meet the online detection requirements.(3)Based on image processing methods,histogram equalization,Gaussian filtering,image segmentation,and morphological operation are utilized to extract the connected component of powder spreading defects.The information of defects can be obtained by connected component analysis to,such as quantity,location,size and area.(4)An image processing-based method and method combining YOLOv3 and image processing for online monitoring of powder spreading defects are established.The two methods are examined using images of 100 layers of spreading(including normal layers and layers with spreading defects).The results show that both methods can be applied in the online monitoring of powder spreading process.Each method has its own advantages and disadvantages,the optimal method must be chosen according to the situation.