Application of Transfer Learning in Metalworking Fluid Distinction

This contribution introduces a Transfer Learning (TL) approach for the diagnostic task to distinguish the ingredients of a typical production machine element: metalworking fluid (MWF). Metalworking fluids are oil or water-based fluids used during machining and shaping of metals to provide lubrication and cooling. Additives in MWF affect their performance in different metalworking processes. Performance evaluation of MWF is of relevance for product development as well as for condition monitoring. In this contribution, for the first time, Transfer Learning is adapted for MWF distinction. Firstly, two experiments are designed to get Acoustic Emission (AE) signals from thread forming processes using variant MWF. In the first experiment, eleven kinds of water-based MWF are applied and AE signals are saved into dataset A, while in the second experiment, other five MWF are used in the process of thread forming and AE signals are stored in dataset B. A convolutional neural network (CNN)-based data mining approach including data segmentation, Short-Time Fourier Transform (STFT) and data normalization algorithms is developed from dataset A. Performance of the proposed approach in dataset A is good. Afterwards, parameters in data processing and hyperparameters in CNN of the approach are transferred into dataset B. Results of dataset B show that Transfer Learning allows suitable MWF distinction.