Recognition and Localization of FBG Temperature Sensing Based on Combined CDAE and 1-DCNN

Abstract

In quasi-distributed fiber Bragg grating (FBG) temperature sensor networks, noise and spectral distortions affect the demodulation accuracy of the fiber gratings. To address this issue, we construct a sensor network using spectral encoding and propose a novel approach that combines convolutional denoising autoencoder (CDAE) and 1-D convolutional neural network (1-DCNN), where CDAE is used for denoising FBG reflection spectra and 1-DCNN is employed for temperature state recognition and temperature demodulation of FBG sensors. The proposed method applies to FBG reflection spectra with different input SNR levels ranging from 0 to 20 dB. Experimental results demonstrate that this CDAE is effective in high-fidelity denoising of the original spectral signals and it outperforms other machine learning techniques. The 1-DCNN model achieves a recognition accuracy of 98.2% for FBG temperature states, with a goodness-of-fit value of 0.9994 for the relationship curve between predicted and actual temperatures, and a root-mean-square error (RMSE) of only 0.3049 °C. This research provides an efficient solution for FBG-based sensor networks.