A Novel Compact Fiber Optic Concentration Sensing System Based on Machine Learning Demodulation

Abstract

A novel compact optical fiber concentration sensing system based on machine learning was proposed and experimentally demonstrated in this paper. The Michelson interferometer (MI) was realized by multiple arc discharge performed on the end of a section of bent bare single-mode fiber (SMF). To improve the stability and accuracy of demodulation, machine learning based on long short-term memory (LSTM) was employed and it provided an accuracy of 97.5%, which is more stable and accurate than conventional peak wavelength tracking due to the fact that LSTM can avoid the effects of dip selection, wavelength sampling rate and spectral noise on the peak wavelength tracking. Furthermore, the proposed sensing system has the advantages of compact size, low cost, high robustness, and ease of fabrication.

Keywords

• Concentration sensing
• fiber optic Michelson interferometer
• machine learning