High-Precision Wide-Range FBG interrogation based on AWG and Ultra-Lightweight Machine learning model

Fiber Bragg Grating (FBG) sensors have gained widespread application in environmental monitoring and industrial fields due to their high sensitivity, simple structure, and miniaturization capabilities. However, high-precision interrogation of FBG sensing signals remains a challenge, as existing systems suffer from limited dynamic range, significant errors, and reliance on costly high-precision instruments, impeding on-chip integration. To address these challenges, we propose an algorithm-photonic co-designed on-chip interrogation system that integrates a customized Arrayed Waveguide Grating (AWG) chip with a polynomial regression-based machine learning model. The AWG chip is optimized through a straightforward design that avoids additional fabrication complexity, while the machine learning algorithm is ultra-lightweight, enables rapid training, and delivers high accuracy. The model was trained on a dataset augmented with Gaussian noise-augmented and validated through a randomized repeat testing approach, ensuring robustness and generalizability. Experimental results demonstrate continuous interrogation over a wavelength range of 1537.4 to 1561.4 nm, achieving a root-mean-square error (RMSE) as low as 0.45 pm to 0.99 pm. This cost-effective, wide-range, and high-precision system demonstrates significant potential for field sensing and FBG sensor networks.