Detection of surface defects on aerospace components in images with Deep Learning
As part of aircraft maintenance, components are visually inspected for surface defects. Here, it must be ensured that all defects are found. An inspection with image recordings enables defect detection with computer algorithms. In addition to image processing, deep learning is a promising approach for detection. In this work, defect detection with Deep Learning is to be investigated using a reference component. Since defects are/were not documented with photos in the context of maintenance, a database with synthetic training images is to be created first, possibly using existing databases. Then a neural network is to be implemented, trained with the training data and tested using a real test setup for inspecting the component with a camera. Among other things, the influence of exclusively synthetic training images and the robustness/reliability of the detection will be evaluated.
Your subtasks
- Building a training database
- Implementation of a neural network for defect detection
- Evaluation of detection in terms of robustness/reliability, influence of synthetic training data
Your profile
- You are interested in machine learning or deep learning
- You have programming skills (Python)
- You have a high degree of initiative
Organisational
- Start date: immediately or November 2020
If you are interested, please contact: Falko Kähler, M.Sc. | f.kaehler@tuhh.de
