Demodulation algorithm for temperature and pressure sensing in side-hole fiber Bragg gratings using a CNN-LSTM network.

The birefringent properties of side-hole fiber Bragg grating(SHFBG) provide an effective solution for the simultaneous measurement of temperature and pressure in confined spaces, such as those encountered in the petroleum and aviation industries. Nevertheless, the combined effects of external pressure and temperature result in the separation and overlap of the reflection spectrum of SHFBG in different polarization directions, posing significant challenges for accurate demodulation using traditional algorithms. In response, we propose a deep learning model based on a CNN-LSTM architecture, specifically crafted to tackle the challenges of high-precision temperature and pressure demodulation in SHFBG. The experimental results indicate that the model performs exceptionally well even in datasets with low wavelength resolution and high signal-to-noise ratio, achieving a root mean square error (RMSE) of only 0.296 MPa for pressure and 0.276°C for temperature, while on the experimental measurement dataset, the errors are 2.425°C and 2.243 MPa. Comparative experiments with various neural network architectures demonstrate that our CNN-LSTM network surpasses other neural network models in terms of training accuracy and generalization capability. This research paves the way for further studies and broader applications of SHFBGs.