eduzhai > Applied Sciences > Engineering >

On Transfer Learning of Traditional Frequency and Time Domain Features in Turning

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: There has been an increasing interest in leveraging machine learning toolsfor chatter prediction and diagnosis in discrete manufacturing processes. Someof the most common features for studying chatter include traditional signalprocessing tools such as Fast Fourier Transform (FFT), Power Spectral Density(PSD), and the Auto-correlation Function (ACF). In this study, we use thesetools in a supervised learning setting to identify chatter in accelerometersignals obtained from a turning experiment. The experiment is performed usingfour different tool overhang lengths with varying cutting speed and the depthof cut. We then examine the resulting signals and tag them as either chatter orchatter-free. The tagged signals are then used to train a classifier. Theclassification methods include the most common algorithms: Support VectorMachine (SVM), Logistic Regression (LR), Random Forest (RF), and Gradient Boost(GB). Our results show that features extracted from the Fourier spectrum arethe most informative when training a classifier and testing on data from thesame cutting configuration yielding accuracy as high as 96. However, theaccuracy drops significantly when training and testing on two differentconfigurations with different structural eigenfrequencies. Thus, we concludethat while these traditional features can be highly tuned to a certain process,their transfer learning ability is limited. We also compare our results againsttwo other methods with rising popularity in the literature: Wavelet PacketTransform (WPT) and Ensemble Empirical Mode Decomposition (EEMD). The lattertwo methods, especially EEMD, show better transfer learning capabilities forour dataset.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×