Comparing Transformer with LSTM recurrent neural networks to predict vibration modes generated by the diffraction pattern

Po-Han Lee^1*, Chen-Hsin Lu¹, Che-Hsien Lin², Chun-Yi Chang¹, Hsiao-Zhu Hong¹, Xin-Han Tsai¹, Anthony An-Chih Yeh¹, Kai-Jong Chou¹

¹Affiliated Senior High School, National Taiwan Normal University, Taipei, Taiwan
²Taipei Fuhsing Private School Taipei, Taipei, Taiwan

* Presenter:Po-Han Lee, email:leepohan@gmail.com

This paper illustrates the prediction of the vibration modes by using transformer with long and short-term memory (LSTM) networks. The neurons networks try to predict vibration mode generated by the handwritten behavior on the optical desk, in which the vibration modes are detected by the diffraction pattern by utilizing the Michelson interferometer. To detect the part of translating the vibration signal, we use transformer and LSTM to identify the data and the model is built with the software tools of Tensorflow and Keras. After the process of training, we successfully predict the diffraction vibration pattern of ripple generated by the behavior of letter handwriting and the accuracy reaches 90% for both methods. This work provides the better results of prediction by using transformer than that of using (LSTM) networks. Such detection of various vibration patterns enables us to understand deeply about the field of the material science in the future.

Keywords: Vibration modes, Michelson interferometer, LSTM, transformer