Chatter Monitoring of Machining Center Using Head Stock Structural Vibration Analyzed with a 1D Convolutional Neural...

Area of Science:

• Mechanical Engineering
• Manufacturing Processes
• Vibration Analysis

Background:

• Real-time chatter detection is essential for maintaining workpiece surface quality and reducing defects in milling operations.
• Existing methods may not fully capture the subtle vibrational changes indicative of chatter.
• Understanding the relationship between structural vibrations and chatter states is key to developing effective detection strategies.

Purpose of the Study:

• To propose a novel methodology for real-time chatter detection in milling based on machine head stock structural vibration.
• To investigate the distinct vibration characteristics of the head stock during stable and chatter milling conditions.
• To develop an effective feature extraction and classification approach for chatter identification.

Main Methods:

• Measurement of machine head stock structural vibration during milling.
• Application of a short-pass lifter to the cepstrum to isolate structural vibration modal components.
• Analysis of vibration magnitude differences for rigid body and bending modes between stable and chatter states.
• Utilization of a one-dimensional convolutional neural network (1D-CNN) for feature extraction and classification from the liftered spectrum.

Main Results:

• A clear dependence of vibration characteristics on cutting states (stable vs. chatter) was identified using the liftered spectrum.
• The vibration magnitude for rigid body modes decreased, while bending modes increased during chatter.
• The 1D-CNN successfully extracted features from the liftered spectrum, enabling accurate demarcation between stable and chatter states.
• The proposed method demonstrated high chatter detection efficiency across various cutting parameters.

Conclusions:

• Machine head stock structural vibration analysis, particularly focusing on modal components, is a viable approach for real-time chatter detection.
• The combination of cepstral liftering and 1D-CNN provides an effective framework for classifying milling states.
• This methodology offers a promising solution for enhancing workpiece quality and process efficiency in milling operations.