Highly sensitive V-groove refractive index sensor based on Box–Behnken design optimization and DNN prediction

A highly sensitive refractive index (RI) sensor based on surface plasmon resonance photonic crystal fiber (SPR-PCF) with a V-groove open channel and hexagonal air holes is proposed in this study. The sensor’s performance is numerically investigated via the Finite Element Method (FEM) implemented in COMSOL Multiphysics. The three structural parameters of polishing depth, gold film thickness, and hole spacing are optimized using the Box–Behnken Design (BBD) response surface method, achieving a maximum wavelength sensitivity of 26,000 nm/RIU and a resolution of $3.85\times 10^{-6}$ RIU. The sensor exhibits a broad detection range covering RI from 1.28 to 1.43. Furthermore, a deep neural network (DNN) model is employed to predict the refractive index by fitting the relationship between the resonance wavelength, loss value, and RI. The model achieves an $R^2$ of 0.998, outperforming the potential of neural networks in modeling complex nonlinear interactions. The proposed sensor structure provides a novel solution for practical applications.