A transfer learning approach for chatter detection in multi-posture robot machining

Zhicong Rong, Ali Khishtan, Jihyun Lee,
A transfer learning approach for chatter detection in multi-posture robot machining,
Manufacturing Letters,
Volume 44, Supplement,
2025,
Pages 1395-1404,
ISSN 2213-8463,
https://doi.org/10.1016/j.mfglet.2025.06.159.
(https://www.sciencedirect.com/science/article/pii/S2213846325001932)
Abstract: Chatter stability prediction is crucial for enhancing machining accuracy and surface quality. However, in robotic machining, variations in the frequency response function (FRF) across different robot postures result in corresponding differences in the stability lobe diagram (SLD), making accurate prediction challenging. Impact testing for each posture is costly and time-consuming. To address this, this paper introduces a transfer learning method based on deep neural networks (DNNs) that enables chatter predictions to be transferred across different postures, thereby reducing the need for large datasets and testing time. First, impact hammer testing is conducted for a specific robot posture to generate the FRF and SLD. The simulated SLD data is then used to pre-train the neural network, enabling it to learn the boundaries and patterns of binary stability classification. Subsequently, a small experimental dataset from another posture, containing only a few dozen samples, is used to fine-tune the network, adapting it for chatter prediction across different postures. Experimental validation shows that the predicted SLDs for various postures align closely with experimentally determined stability limits. The results indicate that, compared to traditional machining learning methods, the transfer learning approach significantly reduces the requirement for training data while achieving high prediction accuracy.
Keywords: Chatter vibration; Stability lobe; Transfer learning; Neural networks; Robotic machining