Abstract
The surface quality of CNC machined workpieces is a critical factor in the manufacturingindustry, reflecting the precision and quality of the entire production process. Traditional
quality control methods are time consuming, error prone, and costly. This thesis aims
to develop a machine learning model to predict surface quality based on vibration data,
thereby improving the efficiency and accuracy of quality control.
The thesis begins with an introduction to CNC technology, engineering fundamentals,
and the importance of surface quality. It explains how vibration data can serve as an
indicator of the milling process’s condition. Following this, the principles of machine
learning and relevant algorithms are detailed.
For model development, vibration data are collected and preprocessed to train a robust
machine learning model. Various algorithms including Logistic Regression, K-Nearest
Neighbours, Support Vector Machines and Random Forest are compared to determine
the best approach to predicting surface quality. The selected model is validated, and its
predictive accuracy is assessed.
The results show that the developed machine learning approach is successful in accurately predicting surface quality. Vibration data alone proved to be sufficient to make
accurate predictions, significantly reducing the need for additional, expensive and complex measurement systems. This enables more efficient monitoring and optimisation of
the milling process. Finally, the practical applications of the results and possible future
research directions are discussed.
| Date of Award | 2024 |
|---|---|
| Original language | German (Austria) |
| Supervisor | Michael Bogner (Supervisor) |