Monitoring of machining process anomalies in diamond turning of Ti6Al4V alloy using transfer learning-based algorithms

Abstract

In-process monitoring of anomalies in Ultra Precision Machining (UPM) can improve and ensure product quality and lower manufacturing costs, which is essentially important for industrial-scale ultraprecision manufacturing. UPM is found to be extremely sensitive towards minute instabilities; if such nascent anomalies are not detected can cause irrecoverable defects. However, the classification of the diamond-turning process through conventional monitoring techniques is challenging. This study explores the use of spectrogram-based deep learning to enable real-time, intelligent process monitoring in UPM. The vibrational signals obtained from machining are transformed into log-spectrogram images. These images obtained during machining allow the rendering of more accurate and richer features of signals, as most of the time domain signal obtained in UPM is susceptible to noise and exhibits several non-linearities. The current approach also uses Transfer Learning (TL) to address the feature selection problem. TL is adopted by using the deep learning (DL) models, which have already been developed for classifying different Images. DL pre-trained networks, including VGG19, ResNet50 and Densenet201, are studied for classifying the anomalies. These TL models are applied to the spectrogram images for the classification of normal and abnormal machining in UPM. Among the TL models, the VGG19 model yields the highest classification accuracy at 90%, which demonstrates the potential feasibility of the TL for monitoring process anomalies in UPM.