Categories
Pages
-

DBIS

StEADyML: Stability Evaluation and Detection for Accurate Dynamics using ML

January 27th, 2025

The aim of this thesis is to design, implement, and evaluate a machine learning-based system for detecting chatter in thin-walled workpieces during machining processes. By leveraging MLOps principles, the system will automate the data pipeline from sensor data acquisition to model deployment, ensuring a scalable and efficient workflow. Additionally, the integration of the system within the Blockchain4DataMarketPlace will enable transparent tracking of the detection process and its results, enhancing the overall trust and accountability in manufacturing environments.

Thesis Type
  • Master
Student
Michal Kovac
Status
Running
Presentation room
Seminar room I5 6202
Supervisor(s)
Stefan Decker
Advisor(s)
Michal Slupczynski
Contact
slupczynski@dbis.rwth-aachen.de

Background and Related Work:

Chatter detection in thin-walled workpieces is a critical challenge in manufacturing, particularly in processes such as CNC machining, where vibrations can lead to poor surface finish and material defects. Traditional methods for detecting chatter rely on manual inspections or simplified algorithms that are often inaccurate or computationally expensive. Recent advancements in machine learning (ML) have shown promise in automating chatter detection using sensor data such as vibrations, acoustic signals, and force measurements. However, implementing these solutions in real-world industrial settings requires robust data pipelines that can handle continuous sensor data streams and facilitate model deployment.

MLOps, a set of practices that unifies machine learning system development and operations, can significantly improve the scalability and automation of ML pipelines. Previous research has focused on integrating MLOps into industrial ML systems, enabling end-to-end automation from data collection to model deployment. However, the application of MLOps for real-time chatter detection in thin-walled workpieces remains underexplored.

Expected Contribution:

This thesis will deliver a prototype system for chatter detection in thin-walled workpieces, combining machine learning and MLOps automation for transparent and efficient manufacturing quality control. The findings will contribute to the growing field of industrial machine learning applications and provide valuable insights into the implementation of MLOps in manufacturing environments. The system’s design and evaluation will also inform future research in the intersection of sensor data, machine learning, and decentralized data management.


Prerequisites:
  • Knowledge of Machine Learning (ML) – Familiarity with ML techniques for time-series data analysis, particularly classification and regression models.
  • Experience with MLOps – Understanding of MLOps principles, tools, and frameworks (e.g., MLFlow, Kubeflow, Docker, CI/CD pipelines).
  • Familiarity with Sensor Data – Understanding of sensor data acquisition methods and preprocessing techniques for time-series data.
  • Programming Skills – Proficiency in Python and familiarity with ML libraries (e.g., TensorFlow, Scikit-learn), as well as MLOps tools.