Multi-modal denoised data-driven milling chatter detection using an optimized hybrid neural network architecture.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Milling chatter, a self-excited vibration, degrades surface quality, tool life, and machining efficiency.
• Existing chatter detection methods struggle with accuracy due to limitations in one-dimensional temporal and two-dimensional image modal information.

Purpose of the Study:

• To propose a multi-modal, data-driven milling chatter detection method using an optimized hybrid neural network.
• To enhance the accuracy and robustness of chatter detection in machining processes.

Main Methods:

• A data denoising model combining Complementary Ensemble Empirical Mode Decomposition (CEEMD) and Singular Value Decomposition (SVD), optimized by the Ivy algorithm.
• Extraction of multi-modal data features using time-frequency domain and Markov transition field methods, with sensitivity analysis via Pearson correlation coefficient.
• Construction of a hybrid neural network (DBMA) integrating dual-scale CNNs, Bi-GRUs, and attention mechanisms, with hyperparameter optimization using the Ivy algorithm.

Main Results:

• Effective denoising of machining signals and the utilization of multi-modal data significantly improved state detection accuracy.
• The proposed DBMA model demonstrated superior stability and robustness compared to existing methods.
• t-SNE visualization confirmed effective feature extraction across different network layers.

Conclusions:

• The proposed multi-modal denoised data-driven approach effectively addresses the limitations of traditional chatter detection methods.
• Optimized hybrid neural networks combined with advanced signal processing techniques offer a promising solution for accurate and robust milling chatter detection.