Artificial neural network based optimization for Ag grated D-shaped optical fiber surface plasmon resonance refractive index sensor

Abstract

This study reports the optimization of fiber optic SPR refractive index sensor parameters with the simulation of finite element method (FEM) and artificial neural network (ANN) model. To demonstrate the applicability of the algorithm, we examined an Ag-grated D-shaped fiber optic sensor configuration with 4 basic input parameters with the aim of reaching the highest sensitivity. Through the conventional optimization, the best parameter set appeared to be a 10 nm air gap distance between the gratings (a), 20 gratings (N), 50 nm residual cladding thickness (d), and 70 nm silver layer thickness (Ag_th) at the indices of 1.35 and 1.39 yielding a sensitivity of 3775 nm/RIU. A close match is found between the actual and predicted sensitivity. 199 input data obtained from FEM are used for training by Leave One Out Cross-Validation (LOOCV) approach with R-squared value of 0.98, and the trained model with R-squared value of 0.97 is implemented in the Genetic Algorithm. We achieved the sensitivity of 3890 nm/RIU at the predicted a, N, d, and Ag_th of 10 nm, 20, 50 nm, and 75 nm, respectively. Future studies may further improve these results by integrating other algorithms. This method may apply to different and more complex structures to observe the correlation between the parameters, cover an entire range of parameters, and get more accurate results, especially with a high number of inputs requiring less time and computing effort. The proposed method carries great potential to improve the sensing ability and bring a new perspective to the literature.