In-Situ Robotic Point Cloud Completion of Additively Manufactured Parts via Dual-Branch CNN Using Partial Measurements and CAD Models

Description

This research data includes the source code and point cloud datasets used for missing region detection and point cloud completion of additively manufactured blisk components. The Classification module implements a multi-stage missing region detection method based on hierarchical clustering. The pressure side and suction side are processed separately by comparing the CAD point cloud with the measured point cloud to identify the missing and non-missing CAD points. The Completion module implements the proposed dual-branch CNN-based point-cloud completion method. The CAD and measured point clouds from the pressure and suction sides are combined to form the complete blisk dataset. Target displacement values are generated from the CAD and measured point clouds and used to train and validate the dual-branch CNN. The trained network is then used to predict the displacement values of the detected missing CAD points and complete the missing regions. The provided MATLAB data files contain the datasets and trained network required to reproduce the validation and prediction procedures. All algorithms were implemented in MATLAB R2024a.

Institutions

• Soochow University

  Jiangsu, Suzhou

Categories

Licence