Please use this identifier to cite or link to this item: https://hdl.handle.net/20.500.11851/10798
Title: A novel transfer learning framework for chatter detection using convolutional neural networks
Authors: Unver, H.O.
Sener, B.
Keywords: Chatter detection
CNN
EEMD
Transfer learning
Convolution
Convolutional neural networks
Deep learning
Intrinsic mode functions
Learning systems
Machining centers
Numerical methods
Time domain analysis
Chatter detection
Convolution neural network
Convolutional neural network
Detection and avoidances
EEMD
High quality
Higher efficiency
Learning frameworks
Regenerative chatters
Transfer learning
Milling (machining)
Issue Date: 2023
Publisher: Springer
Abstract: Detection and avoidance of regenerative chatter play a crucial role in ensuring the high quality and efficiency of machining operations. Predominant analytical approaches provide stability lobe diagrams for machining processes. Deep learning is a general term given to the most recent and successful group of machine learning methods that proved great promise in many areas of human life. This study purposes a novel transfer learning framework that combines analytical solutions and convolution neural network (CNN) under a novel transfer learning framework. Stability lobes and numerical time-domain solutions of analytical methods are used to train and label, arguably one of the most successful CNN architectures, AlexNet. This approach eliminates the need for a time-consuming and costly experimental data collection phase for training. Furthermore, an ensemble empirical mode decomposition based signal pre-processing method is developed. An IMF-based multi-band ensemble approach is proposed where only intrinsic mode functions relevant to each modal frequency of the system are selected based on their entropy increase and used in training multiple AlexNet instances. The measured data were collected during shoulder milling from a CNC-vertical milling machine. The results revealed considerable success in several scenarios ranging from 82 to 100%, without using any experimentally measured data in training. © 2021, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
URI: https://doi.org/10.1007/s10845-021-01839-3
https://hdl.handle.net/20.500.11851/10798
ISSN: 0956-5515
Appears in Collections:Scopus İndeksli Yayınlar Koleksiyonu / Scopus Indexed Publications Collection

Show full item record



CORE Recommender

Google ScholarTM

Check




Altmetric


Items in GCRIS Repository are protected by copyright, with all rights reserved, unless otherwise indicated.