부산대학교 도서관

This work focuses on developing a sound-based anomaly detection model for predictive maintenance in vertical milling machines. The curated dataset includes a wide range of normal and anomalous sound patterns encountered during machine turning operations, with a specific focus on cutting tool wear during the milling process. An autoencoder-based unsupervised machine learning technique is employed to detect anomalies by comparing reconstructed outputs with original inputs. The model is seen to perform better with a longer duration of audio training samples. The results demonstrate the feasibility and efficacy of the system in reducing downtime, improving productivity, and optimizing maintenance practices.