3DK - (1) 3DK - October 23, 2015 (1) PURSUING MECHANICAL PART FEATURE RECOGNITION THROUGH THE...
-
Upload
esmond-powell -
Category
Documents
-
view
215 -
download
0
Transcript of 3DK - (1) 3DK - October 23, 2015 (1) PURSUING MECHANICAL PART FEATURE RECOGNITION THROUGH THE...
3DK3DK - - April 21, 2023April 21, 2023 ( (11))
PURSUING MECHANICAL PART FEATURE RECOGNITION
THROUGH THE ISOLATION OF 3D FEATURES IN ORGANIC SHAPES
IMECE 2002 Conference –New Orleans
Suraj Mohandas
Mark R. Henderson
3DK3DK - - April 21, 2023April 21, 2023 ( (22))
CONTENTS OF THE PRESENTATION
Feature Recognition BasicsImproving Feature RecognitionSegmentation and Classification of Geometric
ModelsResultsConclusionsFuture Goals
3DK3DK - - April 21, 2023April 21, 2023 ( (33))
Introduction
Conventional Feature Recognition of prismatic shapes has had mixed results
Hypothesis: If we can classify complex organic shapes, it may help in classifying simpler engineering shapes
Goal: Develop a method to classify organic shapesResults: Can improve future feature recognition
strategies
3DK3DK - - April 21, 2023April 21, 2023 ( (44))
Common Feature Recognition Strategies
Volume Decomposition Graph-matching Algorithms AI-based and rule-based systems Geometric reasoning Neural nets
Feature Recognition is still an unsolved problem
3DK3DK - - April 21, 2023April 21, 2023 ( (66))
Why Does It Fail ?
Interacting FeaturesWhen Complex real-world
parts are involved hundreds of surfaces lots of curves hundreds of features
From “Feature Recognition for Process Planning: A Status Report - William C. Regli , Mike Pratt Jung Hyun Han”
Organic shapes possess these and other characteristics
3DK3DK - - April 21, 2023April 21, 2023 ( (77))
Putting The Features To Work
DEFINITION: Features are curvature-based regions on a 3D triangulated and subdivided object
GOAL: Extract an overall shape signature
Method: Divide and Conquer Ultimately, re-assemble the
classified regions to recognize complete shapes
3DK3DK - - April 21, 2023April 21, 2023 ( (88))
The Femur Region subdivision is Dependent on a parameter (Threshold) Study to investigate optimal Threshold values for desired Subdivisions What is the desired subdivision ?
3DK3DK - - April 21, 2023April 21, 2023 ( (99))
Generating Regions
STEPS:
Multiple Scanning Merging -> Triangular meshOptional DecimationWatershed (Mangan & Whitaker)
3DK3DK - - April 21, 2023April 21, 2023 ( (1010))
A Little Anatomy
Major Trochanter (Protrusion)
Minor Trochanter
Supracondyloid Fossa (Depression)
Bone organic features are irregular in
nature.
3DK3DK - - April 21, 2023April 21, 2023 ( (1111))
A Little Anatomy
Medial Condyle
Lateral Condyle
3DK3DK - - April 21, 2023April 21, 2023 ( (1212))
A Little Anatomy
Head
NeckFovea Capitus
3DK3DK - - April 21, 2023April 21, 2023 ( (1313))
A Little Anatomy
3DK3DK - - April 21, 2023April 21, 2023 ( (1414))
Mechanical Part
Parts Generated In Acis® Modeler and Then Subjected to
Watershed
3DK3DK - - April 21, 2023April 21, 2023 ( (1515))
Segmentation using Watershed
Segmentation threshold is key in the separation of the object into features
Calculate curvature based on the triangulated model Curvatures available are Principal curvatures,
Gaussian curvature, mean, RMS, absolute, etc. Step 1: Curvature Calculation
3DK3DK - - April 21, 2023April 21, 2023 ( (1616))
Significance Of Threshold
Threshold Defined as the watershed depth Determines number of regions on object
Resulting Region definition is sensitive to the Threshold value
Step – Try Individual threshold values for individual regions
3DK3DK - - April 21, 2023April 21, 2023 ( (1717))
WATERSHED STEPS
Labeling Of Local Minima
Flat Plateau
Flat Minimum
Labeling Flat Region
OR
Bottom Up Approach
Top Down Approach *
OR
Step 2
Step 3
3DK3DK - - April 21, 2023April 21, 2023 ( (1818))
WATERSHED MERGING (Final Step)
Watershed Depth
Vertex Of Current Region
Vertex Of Neighboring Region
Merging Adjacent Regions With Shallow Depths
Threshold
3DK3DK - - April 21, 2023April 21, 2023 ( (1919))
Effect Of Threshold
0.001 0.025 0.1 0.5
3DK3DK - - April 21, 2023April 21, 2023 ( (2020))
Results & Conclusions
Feature Snapshot Threshold Value
BEST
Head & Fovea
Capitus
0.14945 – 0.19990
WORST
Trochantic Fossa
0.11301- 0.12246
3DK3DK - - April 21, 2023April 21, 2023 ( (2121))
Best and Worst Subdivision Results
BEST RESULT: Good correlation to the actual anatomical feature
for a given threshold range.
WORST RESULT: Poor correlation to the actual anatomical feature at
any threshold
3DK3DK - - April 21, 2023April 21, 2023 ( (2222))
Results & Conclusions
Feature Snapshot Threshold Value
Neck 0.26127 - 0.27894
Minor
Trochanter
0.05331 - 0.05907
3DK3DK - - April 21, 2023April 21, 2023 ( (2323))
Results & Conclusions
Decimation changes regions loss of important curvature data granularity.
Some features were more easily picked up than others Some feature segmentations (e.g., large depressions - fossae)
less sensitive to threshold variation. Difficult to predict the optimal threshold – it depends on the
segment desired Threshold range can be predicted for similar features on two
similar types of objects Can predict good threshold ranges through experience
3DK3DK - - April 21, 2023April 21, 2023 ( (2424))
Using the Segments to Create the Feature Signature
Array of Single Value Scale & Orientation Invariant Parameters
calculated from the region geometry
Metric 1 Metric 2 Metric 3 Metric (n-1) Metric n
Assuming a good subdivision, for each region calculate a Shape Signature
Shape Signature Vector Definition:
3DK3DK - - April 21, 2023April 21, 2023 ( (2525))
Future
Develop method to predict optimal feature extraction thresholds for desired features.
Predict decimation limitCluster shape signatures to classify shape
3DK3DK - - April 21, 2023April 21, 2023 ( (2626))
Acknowledgements
The Students and Faculty in the “Partnership for Research In Spatial Modeling”- PRISM at Arizona State University
Myungsoo Bae (PhD Candidate ASU) – Interface design