Acoustic emission reflection signal classification of PVDF-type AE sensor using convolutional neural network-transfer learning

This study proposes a polyvinylidene fluoride (PVDF)-type AE sensor to demonstrate the feasibility of replacing conventional acoustic emissions (AE) sensors. The Hsu-Nielsen test was used to generate the signals, and conventional AE and PVDF-type AE sensors were used to sample the signals. To verify that the PVDF-type AE sensor can classify different characteristics, direct wave signals and signals distorted due to specially designed settings were collected and analyzed. For effective data processing, a convolution neural network (CNN) was constructed and trained with AE spectrogram images after wavelet packet transform (WPT) from both AE sensor signals and PVDF-type AE sensor signals. The results of CNN–WPT showed that direct and indirect waves can be distinguished using PVDF-type AE sensor signals with almost the same accuracy as conventional AE signals. To improve the accuracy of the classification, transfer learning was used to increase the accuracy of the validation and reduce training time. This demonstrates that PVDF-type AE sensors can replace AE sensors when acquiring and classifying AE signals.