Self-Attention AI Model for Practical Sensor Networking: Demodulation of Long-Period Fiber Grating Sensor Cascaded With FBG Sensor Array

This paper proposes a cost-effective solution for interrogating long-period fiber grating (LPFG) sensors using an optical fiber sensor network. The network comprises an array of fiber Bragg grating (FBG) sensors, whose reflection spectrum are modulated by the LPFG’s transmission spectrum. A self-attention AI model decodes these spectral modulations to extract the LPFG’s resonant wavelength. The AI model dynamically filters the FBG reflection spectra, focusing on the most relevant features for LPFG demodulation. It is trained using synthetic data and validated with measured LPFG spectra. The system achieves sub-nanometric demodulation resolution (0.75 nm RMSE and 0.0335% MAPE) for a set of 73 LPFG spectra. In a practical application using a refractive index LPFG sensor, the system demonstrates a relative error of less than 1%. This approach, combining an FBG array with a synthetically trained self-attention AI model, offers a promising pathway towards more practical and cost-effective LPFG interrogation in multi-sensor networks.