Highly sensitive vibration measurement method of nanomechanical resonators using a convolutional neural network

Abstract

Abstract Nanomechanical resonators can detect various small physical quantities. It is necessary to increase the sensitivity of nanomechanical resonators. In this study, a deep learning-based method is developed for sensitive detection of vibration states of nanomechanical resonators. It enables one to explore the most highly sensitive vibration states without missing targets while conventional methods focus only on the resonant frequency. The proposed method was evaluated with laser intensity. Sensitive measurement of laser intensity has the potential for applications in laser wavelength measurement. The proposed method could identify laser intensity at intervals of 0.03 mW at which the resonance frequency did not shift. Grad-CAM was used to clarify features on the vibration spectra used in the convolutional neural network. Frequency bands other than the resonance frequency band were used for the laser intensity measurement. This result implies that there is a possibility that highly sensitive vibration behavior exists outside the resonance frequency.