Description

Optical quality control of industrial steel bands in steel rolling works by detecting the frequency and type of defects that occur on the steel band at each inspection station in order to assess the overall quality of the product and identify emerging problems in production. Changes in the conditions under which the observations are made, such as novel lighting conditions or changes to the camera position, can lead to detections that the computer vision model fails to classify correctly. In addition, defects that belong to classes that could not be represented in the training data may occur during deployment. The goal of this project is to test methods for out-of-distribution detection and anomaly detection on the classifier in a computer vision system for steel band quality control in order to detect domain drift and ideally isolate individual anomalies in incoming data in a reliable manner.
To this end, the first phase of the project is to assess the existing data and the target application domain with respect to the types and characteristics of domain drifts and novel defects. The outcome will be a testbed to evaluate the performance of drift and anomaly detectors by simulating the real-world divergence of deployment data from training data in the domain of optical quality control of industrial steel bands.
In the second phase, the testbed will be used to define and test a process for selecting and calibrating drift and anomaly detectors based on the training data available on-site. The target of this phase of the project is to reliably monitor the health of the defect detection system and to alert when incoming data indicates a drift in operating conditions that the computer vision model is unable to handle or that requires human intervention.

This Thesis is conducted and supervised by Prof. Dr. Laurenz Wiskott and Pavlos Rath-Manakidis from the Institut für Neuroinformatik (www.ini.rub.de) in collaboration with IMS Messsysteme GmbH (www.ims-gmbh.de) and takes place as a 6-month paid internship at IMS Messsysteme GmbH in 42579 Heiligenhaus.
Please apply to pavlos.rath-manakidis@ini.rub.de and include your transcript of records and a CV without a picture.

Keywords: Anomaly Detection, Optical Quality Assessment, Computer Vision, Steel Rolling, Machine Learning, Pipeline Development.

Prerequisites:

• Experience with the Python programming language.
• Completed courses, internships or projects on artificial intelligence or machine learning. Preferably in the field of computer vision.
• Willingness to work on site at IMS Messsysteme GmbH in 42579 Heiligenhaus.

Main References:

• Salehi, Mohammadreza, Hossein Mirzaei, Dan Hendrycks, Yixuan Li, Mohammad Hossein Rohban, and Mohammad Sabokrou. 2022. “A Unified Survey on Anomaly, Novelty, Open-Set, and Out-of-Distribution Detection: Solutions and Future Challenges.” arXiv. http://arxiv.org/abs/2110.14051.
• Han, Songqiao, Xiyang Hu, Hailiang Huang, Minqi Jiang, and Yue Zhao. 2022. “ADBench: Anomaly Detection Benchmark.” Advances in Neural Information Processing Systems 35 (December): 32142–59.
• Yang, Jingkang, Pengyun Wang, Dejian Zou, Zitang Zhou, Kunyuan Ding, Wenxuan Peng, Haoqi Wang, et al. 2022. “OpenOOD: Benchmarking Generalized Out-of-Distribution Detection.” Advances in Neural Information Processing Systems 35 (December): 32598–611.

Additional References:

• Akcay, Samet, Dick Ameln, Ashwin Vaidya, Barath Lakshmanan, Nilesh Ahuja, and Utku Genc. 2022. “Anomalib: A Deep Learning Library for Anomaly Detection.”
• Zhao, Yue, Zain Nasrullah, and Zheng Li. 2019. “PyOD: A Python Toolbox for Scalable Outlier Detection.” Journal of Machine Learning Research 20 (96): 1–7.
• Céspedes-Sisniega, Jaime, and Álvaro López-García. 2023. “Frouros: A Python Library for Drift Detection in Machine Learning Systems.” arXiv. https://doi.org/10.48550/arXiv.2208.06868.
• Malinin, Andrey, Liudmila Prokhorenkova, and Aleksei Ustimenko. 2021. “Uncertainty in Gradient Boosting via Ensembles.” arXiv. https://doi.org/10.48550/arXiv.2006.10562.