Hybrid transfer learning framework for chatter mark detection on a 20-high cold rolling mill

Datasets underpinned a study, "A domain adaptive hybrid learning framework for vibration-based chatter detection on a 20-high rolling mill." This work aimed to apply a hybrid transfer learning framework for chatter mark detection on a 20-high cold rolling mill. The framework makes use of both a physics-based dynamic model and a data-driven model. The physics-based model is used for data augmentation by generating large synthetic datasets. The synthetic datasets are used along with limited experimental data acquired from the actual asset. The framework applies transfer learning as a tool to calibrate a CNN-based classifier to generate descriptive and domain-invariant features. This enables effective generalization of the information in the synthetic domain that can be extrapolated to a given experimental domain, alleviating real-world data requirements.