HOW TO RECONSTRUCT DAMAGED PARTS BASED ON PRECISE AND (PARTLY-)AUTOMATISED SCAN METHODS cirp GmbH 3D models features spare part identification load simulation mesh repair search in data bases machine learning images Cloud 3D data Mobile 3D-Erfassung und 3D-Druck für industrielle Anwendungen M3D
Process chain 2
Challenges Scanning is an optical method! Dirty, rusty Shiny and reflecting Transparency Damaged Hidden areas Assembly golfball gear knop transparent part lightweight part chrome strip armrest 3
Scanning method With photos Mobile Online feedback required => further research activity Still limited accuracy Step by step with structered light => state of the art equipment available best accuracy with tripod => needs educated user and doesn t fit into trouser pocket With structured light automised by robot Fast Economical reliable 4
Motivation Mobile automated scanning station 3D-Scanning of arbitrary unknown objects requires manual work and expert knowledge is tedious and time-consuming, especially repositioning of scanner or object Potential for massive time saving Advances in COBOTS Light-weight collaborative robots Advances in camera and laser technology High-resolution and high-frequency 3D-scanners project preparation scan part data reconstruction sum 1. scan 2. scan 3. scan sum 1. scan 2. scan 3. scan sum cleaning cleanup Geomagic Solidworks sum golfball gear knob 10 min 10 min - 20 min 10 min 15 min - 25 min - 5 min 45 min 15 min 60 min 110 min golfball gear knob (2) 10 min 5 min - 15 min 10 min 10 min - 20 min 10 min 5 min 45 min 15 min 60 min 110 min chrome strip 20 min 10 min 30 min 80 min 40 min 120 min 10 min 5 min 120 min 270 min lightweight part 10 min 5 min - 15 min 30 min 15 min - 45 min 10 min 15 min 30 min 30 min 60 min 145 min transparent part 10 min 5 min - 15 min 50 min 10 min 10 min 70 min 25 min 5 min 150 min - 150 min 265 min project golfball gear knob golfball gear knob (2) chrome strip lightweight part transparent part Combination through software to automate the scanning task Feedback loop between scanner and robot through view planning sum 110 min 110 min 270 min 145 min 265 min target time saving through scan automation MOVE target time saving through automation over 50% and more 25% and more SCAN PLAN 5
Setup Mobile automated scanning station Camera Image Data View Planning Windows PC USB Motion Planning Linux PC TCP/IP Microcontroller Joint Values / 125 Hz Camera Trigger at 20 Hz UART UR5 Robot Turntable 6
405nm Laser-based 3D-Scanner Method of operation Triangulation displacement of the line observed in the image allows to measure distance Advantage: Robustness simple structured pattern one line, less confusion high laser intensity to avoid spraying dark or shiny objects bandpass filter to block other distracting light sources Disadvantage and Solutions low information value per image use higher frequency external localization needed to merge lines to a 3D surface scan use robot coordinates Line laser Without filter Bandpass filter Monochrome camera With filter 7
System Calibration and View Planning Laser scan: Initial automated system calibration procedure 1. Intrinsic Camera Calibration (Camera Lens Distortion) 2. Extrinsic Scanner Calibration (Laser to Camera) 3. Hand-Eye Calibration (Scanner to Robot) Real-Time Synchronization of Robot and Scanner Microcontroller for triggering and transferring coordinates Outlook on further automatization tasks: View Planning: Identify holes, resolve occlusions and plan next best scanner views for automatic scan completion Registration: Automatic registration of partial scans after manual repositioning on turntable Point cloud Analyzed Mesh 8
Potential defects 9
Repair on mesh level Why? 3D-printers and geometry processing algorithms usually need watertight meshes: No holes No self intersections Manifold geometry (locally every part of the mesh is two dimensional) 10
Repair on mesh level Fixing these issues by local manipulation of the mesh is generally hard. We use a global approach: a) Input geometry contains issues such as holes. b) Represent geometry implicitly by the distance to the surface. c) Repair implicit representation by selective smoothing. d) Extract surface as the zero-set using marching cubes. Result is guaranteed to be watertight. unreliable distance value negative distance value positive distance value (a) (b) (c) (d) 11
Repair on mesh level Original scan data containing holes. Reconstructed surface. Holes have been filled in as smooth as possible. 12
Repair on mesh level Original scan data containing holes. Reconstructed surface. Holes have been filled in as smooth as possible. Problem: By smoothing the distance function edges are distorted. Possible solution: Anisotropic smoothing. 13
Repair by reconstruction Why? Achieve proper prametric CAD-Modell and compensate scan deviations by correct geometries 14
Repair by reconstruction broken domes Reconstructed scan data as proper CAD model, with repaired domes Scan data from structured light scanner GOM Processed scan data with proposed identification of regions Colours show deviation from original scan 15
The end questions? 16