Enhanced self-organising map model for surface reconstruction of unstructured data
Surface reconstruction (SR) is a process of recovering the digital representation of an object in reverse engineering. When the unstructured data are applied in the SR process, incorrect surface is produced because the data do not have any connectivity information. Self-Organising Map (SOM) models w...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2023
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/6252/1/CEA_2023_YCC.pdf http://eprints.utar.edu.my/6252/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Surface reconstruction (SR) is a process of recovering the digital representation of an object in reverse engineering. When the unstructured data are applied in the SR process, incorrect surface is produced because the data do not have any connectivity information. Self-Organising Map (SOM) models were proposed to organise the unstructured data to regain the connectivity information, but incorrect surface with holes, internal neurons and different grid sizes problems were appeared. Although the SOM model can generate the correct surface, its output is not in the standard format of Computer Aided Geometric Design. So, Non-Uniform Rational B-Spline (NURBS) surface approximation approach was applied to the output using parameterisation methods. However, the surfaces generated still contain gaps and were not optimal. Hence, the surfaces can be optimised using optimisation techniques. Therefore, the objectives of this research are to propose a SOM model for organising the unstructured data and to present and optimise the NURBS surface approximation approach with Genetic Algorithm (GA), Differential Evolution (DE) and Particle Swarm Optimisation (PSO). The data set used includes four primitive objects and a medical image data. The codes were developed using Microsoft Visual Studio 2022 with C++ programming
iv
and GNUPlot was used to visualise the results. The results shown that the Double Net SOM (DNSOM) model performed faster than 3-D SOM and Cube Kohonen SOM (CKSOM), achieved the lowest Topographic Error and generated the correct surface with fewer neurons compared to CKSOM. Additionally, the improved NURBS approach with Chord Length method was able to generate the correct surface with no gaps and the least surface error. DE can optimise the improved NURBS surface better compared to GA and PSO by achieving 243 out of 280 least optimised surface errors. The research outcomes can be utilised in reverse engineering to recover the surface of an object. |
|---|