Related Experiment Video
Updated: Aug 30, 2025
09:41
Magnet Assisted Composite Manufacturing: A Flexible New Technique for Achieving High Consolidation Pressure in Vacuum Bag/Lay-Up Processes
Published on: May 17, 2018
13.6K
Mask-Point: Automatic 3D Surface Defects Detection Network for Fiber-Reinforced Resin Matrix Composites
Helin Li1, Bin Lin1, Chen Zhang2
1School of Mechanical Engineering, Tianjin University, Tianjin 300072, China.
Polymers
|August 26, 2022
Summary
A new Mask-Point network accurately detects 3D surface defects in fiber-reinforced composites. This advanced system offers superior performance compared to human inspection and other methods.
Area of Science:
- Materials Science
- Computer Vision
- Artificial Intelligence
Background:
- Surface defects in fiber-reinforced resin matrix composites (FRRMCs) negatively impact their aesthetic and functional properties.
- Accurate and efficient detection of these defects is crucial for quality control and performance assurance.
Purpose of the Study:
- To propose a novel, lightweight, two-stage semantic segmentation network, termed Mask-Point, for accurate and efficient 3D surface defect detection in FRRMCs.
- To develop and evaluate the performance of the Mask-Point network on a comprehensive dataset of FRRMC surface defects.
Main Methods:
- Developed Mask-Point, a two-stage network featuring 3D region proposal extractors (RPEs) and a 3D aggregation stage with shared classifier, filter, and non-maximum suppression (NMS).
- Created a new 3D surface defects dataset for FRRMCs comprising approximately 120 million points for training and testing.
- Conducted comparative experiments against other 3D semantic segmentation networks and human inspection.
Main Results:
- Mask-Point achieved high accuracy (0.9997) and mean intersection over union (mIoU) (0.9402) with an inference speed of 320,000 points/s.
- Performance metrics improved with an increasing number of 3D RPEs, though inference speed decreased.
- Mask-Point demonstrated superior segmentation performance over existing methods, with mIoU approximately 30% higher than PointNet.
- A distributed detection system based on Mask-Point outperformed skilled human workers in detecting surface defects on real FRRMC products.
Conclusions:
- The proposed Mask-Point network provides an accurate and efficient solution for 3D surface defect detection in FRRMCs.
- Mask-Point offers a promising approach for defect detection in similar materials, enhancing quality control and manufacturing processes.
- The developed system integrates advanced AI for practical industrial applications, surpassing human capabilities in specific defect detection tasks.
