Intelligent chatter detection in turning operations with imbalanced data using conditional generative adversarial networks

Turning is one of the oldest and most frequently used machining methods in history. Chatter is an undesirable type of vibration caused by the forces between the cutting tool and the work piece. Chatter vibrations, which are the most destructive vibration type for machines, can damage the work piece, cutting tool and machines during the machining process. Despite recent advances in smart data-driven chatter detection methods in turning, most studies assume balanced training datasets for different class conditions. However, chatter signals are often difficult and expensive to collect, resulting in unstable training datasets. In this thesis, in addition to real training data, a deep learning-based chatter detection method is proposed to eliminate data imbalance problems. For the first time, one-dimensional conditional generative adversarial networks are used in chatter detection in order to eliminate the lack of data in the datasets, especially the chatter data. Turning data is obtained with the help of a sensor that can be easily integrated into the system. The CEEMDAN decomposition algorithm is used for the first time within the scope of chatter detection studies in order to eliminate the noise effects in the data collected within the scope of the experiments. The reliability of synthetic data produced in different scenarios is investigated. In the light of the results, the contributions of the synthetic data and the CEEMDAN algorithm to the detection of chatter are presented.