Computer Vision-enhanced In-situ Surface Topography Measurement with Focus Variation Microscopy for Material Extrusion-based Additive Manufacturing

The focus variation microscopy like Alicona G5 performs surpassingly in as-built surface quality analysis for material extrusion-based additive manufacturing yet fails in in-situ scenarios. To enable cheaper and more accessible cameras to obtain surface topography measurements similar to those of Alincona G5, this study proposes a computer vision-enhanced method with a refined window sliding technique for 2D image and 3D point cloud registration to transfer the ability of Alicona G5 to in-situ applications. Deep learning models based on VGG11 and ResNet18 were proposed to establish the relationship between Alicona G5 images and point clouds and XIMEA CCD camera images. A case study of Fused Deposit Modelling manufactured PLA surface predictions showed an average Chamfer Distance of 0.012, Earth Mover Distance of 1.002, Pearson Correlation of 0.985, and relative error of 0.012 against the real surface point cloud with an error of surface roughness prediction of was 17.7% in 19.8 seconds. The results proved that the proposed method held the feasibility to narrow the gap between the effectiveness and efficiency of applying focus variation microscopy, pointing out a cheaper way for in-situ measurement for material extrusion-based additive manufacturing.