Fiber-optic temperature sensor based on beat frequency and neural network algorithm

A multi-longitudinal mode fiber laser sensor (MLMFLS) system based on neural network (NN) algorithm is proposed. Gaussian process regression (GPR) is used to denoise the beat frequency signal (BFS). The frequency change of BFS has a linear relationship with the applied external temperature change. Neural network algorithm is used to fit the linear relationship between the frequency of BFS and external temperature change. Processing and analysis of experimental data showed that the method fitted very well with the relationship between the beat frequency signal and external temperature change. The MATLAB App Designer tool is used to display the wide band frequency spectrum, frequency of BFS and corresponding temperature change. The sensitivity of single beat frequency is 5.204 kHz/℃, the maximum absolute error is ± 0.15 ℃ and the average error is 0.072 ℃. The five beat frequencies sensitivity is 5.206 kHz/℃, the maximum absolute error is ± 0.09 ℃ and the average error is 0.055 °C. Through the combination of neural network algorithm and beat frequency sensing system, simple, fast and real time temperature sensing is realized.