Spectral Demodulation of Fiber Bragg Grating Sensor Based on Deep Convolutional Neural Networks
Zihan Cao, Shengqi Zhang, Titi Xia, Zhengyong Liu, Zhaohui Li
Abstract
This paper presents a new method of demodulating the spectrum of fiber Bragg grating (FBG) based sensors by employing deep convolutional neural networks (DCNN). As a proof of demonstration, FBG-based temperature sensor was utilized to conduct temperature measurement and over 1700 samples of the spectral raw data were recorded to train and validate the DCNN model. Using such method, the temperature information can be directly extracted from the experimentally obtained FBG spectra without any peak tracking algorithms. Since it makes full use of the information containing the full spectrum rather than only the central wavelength, it overcomes the limit of traditional fitting method and could improve the measurement accuracy of FBG effectively, which can reach 99.95% and its mean square error (MSE) is just 0.1080 °C, an order of magnitude less than that achieved by the traditional maximum peak method. The proposed method could reduce the need of high-performance hardware of equipment, whose accuracy can still maintain a high level when the sampling rate is reduced. Additionally, the universality of the method was experimentally demonstrated through the accurate demodulation of tilted FBG spectrum, and the relevant measurand can be retrieved directly from the entire spectrum instead of detecting the change of particular peaks. The proposed approach provides a cost-effective solution for the FBG based sensing system, and is promising for establishing sensing networks to implement smart monitoring.