Research
Automated Defect Detection for Industrial Quality Control
Motivation
The use of machine learning and deep learning in industrial, image-based quality control holds enormous potential, but is currently prevented by two circumstances:
- Supervised learning methods require large amounts of data, including data on faulty states, which is expensive to collect.
- There are no methods suitable for managing the product life cycle of the methods developed.
The methods and current developments in the field of open set recognition offer possible solutions here and are being investigated in this project.
Scientific Questions
To avoid the problem of large data volumes, anomaly detection algorithms are being researched and developed further. In contrast to classical supervised learning methods, the aim of anomaly detection is a descriptive and discriminative characterisation of the normal state.
The algorithms are developed on the public MVTec data set and are also adapted to particular use cases such as defect detection in complex textiles (AiF project OnLoomPattern).
Methods of out-of-distribution (OOD) detection are being researched and developed further as a basis for product life cycle management methods. Combined with explainable AI (XAI) algorithms, they allow both an automatic assessment of whether a machine learning system may be used (OOD) and a verification of that automatic assessment (XAI). The algorithms are further adapted to the particular use case of forgery-resistant animal hair fibre identification (AiF project KiT).
Thesis Topics
New theses in the area of automated defect detection for industrial quality control are advertised regularly. In addition to the Overview , there are also numerous topics that have not yet been advertised, which will be gladly presented in a personal conversation.
Partners
- Prof. Dr. Thomas Gries, Chair of Textile Machinery, ITA – Institut für Textiltechnik, RWTH Aachen University
- Prof. Dr. Andreas Herrmann, Deputy Scientific Director, DWI – Leibniz Institute for Interactive Materials, Aachen
External Funding
- AiF – Forschungsnetzwerk Mittelstand, “Automatic real-time defect detection in textiles with complex patterning structures- OnLoomPattern“ Projektnr. AiF 19811 N
- AiF – Forschungsnetzwerk Mittelstand, “Integrative measuring system for defect detection during textile fabric production”, project no. AiF 355 Z
- AiF – Forschungsnetzwerk Mittelstand, „KIT – Künstliche Intelligenz zur automatisierten, fälschungsresistenten Bestimmung von Tierhaaren“, Projektnr. AiF 21376 N
