Master Thesis - Exploring Anomaly Detection Methods in Time-Series Industrial Data

Project Description:
In modern industrial environments, the ability to detect anomalies in production processes is critical for ensuring operational efficiency, reducing downtime, and maintaining product quality. This project aims to develop and evaluate machine learning models for detecting anomalies in time-series data collected from industrial production systems.

Methodology:
- Literature Study: Review state-of-the-art anomaly detection techniques applicable to time-series industrial data.
- Data Collection and Preprocessing: Acquire and preprocess time-series from industrial sensors or production logs, including handling missing values, normalization, and feature engineering.
- Model Development: Implement and train several selected machine learning algorithms for anomaly detection.
- Unsupervised machine learning algorithms will be the main scope in this project, such as autoencoders, isolation forest, DBSCAN, GMM, etc.
- If we still have additional time, it is also possible to further explore supervised machine learning algorithms. Reinforcement machine learning is out-of-scope in this project.
- Evaluation and Benchmarking: Evaluate model performance using appropriate metrics such as precision, recall, F1-score, and ROC-AUC. Compare the models to determine their strengths and weaknesses in different contexts.
- Visualization and Interpretation: Develop visual tools to interpret model outputs and support decision-making for industrial stakeholders.
- Deployment Considerations: Discuss the potential for real-time deployment and integration into existing industrial monitoring systems.

Expected Outcomes:
- A comparative analysis of different anomaly detection methods on industrial time-series data.
- A robust and interpretable anomaly detection framework suitable for real-world industrial applications.
- Insights into the challenges and best practices for applying machine learning in industrial settings.
 

Candidate Requirements:
- Proficiency in Python programming, particularly in the context of data analysis and machine learning
- Students with backgrounds in data science, computer science/ AI, engineering, technology or applied mathematics are preferred.

Contact information:
If you have questions, please feel free to reach out to Yuxiao Zhao (yuxiao.zhao@tetrapak.com).

We are looking forward to your application, the goal is to start the project fall 2025, apply today – the selection will be continuous.