Point cloud processing - scanning, analysing and validating NPL Freeform Centre of Excellence launch -1 st October Andy Robinson
Contents Heat shield tile measurements Laser scanner performance analysis Freeform software validation Point cloud processing 2
Contents Heat shield tile measurements Laser scanner performance analysis Freeform software validation Point cloud processing 3
Problem Premature failure of heat shield tiles Heat shield tile Npower Gas turbine engine Point cloud processing 4
The approach Scanning using Faro Platinum arm and v3 laser scanner Registration using Polyworks IMAlign Point cloud processing 5
Producing a reference Polygonal mesh Point cloud processing 6
Alignment & comparison Aligned on locating feature Plane, vector, point alignment Point cloud processing 7
Results ±10mm Tile 138 ±5mm Tile 139 Tile 137 Tile 138 Tile 139 Tile 140 Tile 134 Tile 145 Point cloud processing 8
Outcomes Digital inspection gave a good insight into what had gone wrong The deformation was worst in one corner Rotation of the tiles in the engine may have exposed one corner Point cloud processing 9
Contents Heat shield tile measurements Laser scanner performance analysis Freeform software validation Point cloud processing 10
The problem Variability in scanner performance due to Operator Laser power Distance to subject Surface finish Ambient lighting Environment Uncertainty & traceability Point cloud processing 11
The approach Point cloud processing 12
Operator influences 0.12 Standard Deviation [mm] 0.10 0.08 0.06 0.04 0.02 0.00 A B C Operator 0.07 0.06 Standard Deviation [mm] 0.05 0.04 0.03 0.02 0.01 0.00 Software 1 (A) Software 1 (B) Software 2 Point cloud processing 13
Distance influences 0.40 0.35 Standard Deviation [mm] 0.30 0.25 0.20 0.15 0.10 0.05 0.00 600 mm 725 mm 925 mm 1262 mm 1553 mm 2078 mm Point cloud processing 14
Other influences 0.050 0.045 0.040 no coating Shiny mode coating Standard Deviation [mm] 0.035 0.030 0.025 0.020 0.015 0.010 0.005 0.000 1 2 3 4 5 Number of scans Point cloud processing 15
Laser power laser power setting A laser power- setting B Point cloud processing 16
Outcomes Demonstrates the variability in performance Operator often guided by instinct How do we select an appropriate measurement technique? Need for research, standards & validation to give confidence to industry Point cloud processing 17
Contents Heat shield tile measurements Laser scanner performance analysis Freeform software validation Point cloud processing 18
The problem Production Inspection Instrument Software Design Model Part Point Inspection cloud result Error Uncertainty Point cloud processing 19
The objectives To investigate the following functions Alignment Comparison Shape fitting Collaboration with NPL s Mathematical Techniques for Metrology Group Testing Industry wide point cloud processing software packages Point cloud processing 20
The approach Algorithms for data generation & fitting generate CAD models in Matlab. reference datasets are generated using null space method Perturbations applied, along direction of surface normal. For reference data set, compare reference results with test results. standard deviations for inspection geometric parameters for shape fitting > 100 datasets (x,y,z s), produced and analysed Point cloud processing 21
Datasets Type s, standard (no translation or rotation) e, extended (translation + rotation) x, e + random ordering of points Perturbations fine, 0.001 mm standard deviation medium, 0.01 mm standard deviation course, 0.1 mm standard deviation Point cloud processing 22
Basic shapes CAD models Point cloud processing 23
Results Mostly the differences in std deviations were below 1% for more than 98% of datasets For two test cases (with cylinder), the difference was more than 10%. Observed differences are due to the best fit alignment process (default convergence limit) Shape fitting for supported shapes was accurate to <0.005% Point cloud processing 24
Observations Can t always access the full dataset proprietary information Some software tessellates the CAD model can lead to small errors Difficult to analyse the alignment process independently multiple solutions Point cloud processing 25
Next steps Produce CAD models and datasets for industrial objects/shapes turbine blade Repeat exercise with other software packages Point cloud processing 26
Acknowledgements NPL Dr Mike McCarthy, Stephen Brown, Anthony Evenden Prof Alistair Forbes, Minh Hoang Loughborough University Dr Geoff West, Scott Newman Conservation Technologies, National Museums Liverpool Annemarie La Pensee, Joe Parsons Point cloud processing 27
Thank you for your attention!