Deep learning for milling chatter detection: integrating spindle awareness in time series lightweight adaptive network

Date

2025-4-7

Author

Bahçeli, Furkan

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

310
views

0
downloads

Cite This

This thesis introduces the Spindle-Aware Time Series Lightweight Adaptive Network (SA-TSLANet) for improved chatter detection in milling operations. Chatter, a self-excited vibration that degrades machining quality and reduces tool life, remains challenging to detect reliably across varied conditions. The proposed model integrates physics-informed spectral processing with deep learning through the novel Spindle-Aware Adaptive Spectral Block (SA-ASB), which effectively separates spindle harmonics from chatter vibrations. A multi-objective loss function incorporating crest factor, spindle, and ratio components guides the model to learn critical signal characteristics. Experimental validation demonstrates exceptional performance with 0.99 accuracy, 0.98 precision, 0.99 recall, and 0.99 F1 score. SA-TSLANet detects chatter onset 0.19 seconds earlier than traditional FFT methods, providing crucial time for intervention. The model maintains robust generalization (0.98 F1 score) even with limited training data and shows computational efficiency suitable for real-time industrial implementation. The research also explores training exclusively with healthy cutting data to reduce implementation costs. This physics-informed deep learning approach advances machining process monitoring by balancing theoretical rigor with practical applicability, providing a foundation for intelligent manufacturing systems that enhance productivity, quality, and tool utilization.

Subject Keywords

Machine learning, Chatter detection, Signal processing, Time series analysis, Manufacturing automation

URI

https://hdl.handle.net/11511/114633

Collections

Graduate School of Natural and Applied Sciences, Thesis