Objective Fiber Bragg Gratings (FBGs) are widely used in structural health monitoring, aerospace engineering, and other fields due to their advantages of high sensitivity, immunity to electromagnetic interference, and multiplexing capability. However, in long-term practical applications, FBGs are susceptible to stress fatigue, thermal aging, microstructural deterioration, and other environmental stresses. These factors often lead to reflection spectral distortion, including spectral broadening, asymmetry, intensity attenuation, and overlapping, which significantly complicates wavelength demodulation and reduces sensing reliability. To address these challenges, this study proposes an adaptive demodulation method that integrates spectral pattern recognition with the Starfish Optimization Algorithm (SFOA).

Methods The mechanisms behind typical spectral distortion types are first analyzed through numerical simulations. Based on the generalized asymmetric Gaussian model, theoretical expressions for representative distorted spectra are constructed, enabling the wavelength demodulation task to be reformulated as a nonlinear function-fitting optimization problem. To automatically and accurately identify different distortion categories, a convolutional neural network (CNN) is designed and trained using multiple types of distorted FBG spectra, thereby improving the adaptability of the demodulation framework to complex and variable spectral shapes. Following distortion classification, SFOA is applied to optimize the parameters of the spectral model. By mimicking starfish predation and regeneration behaviors, SFOA performs global search within the solution space to obtain the optimal parameter set that minimizes the fitting error, allowing reliable demodulation even under severe distortion conditions.

Results and Discussions Experimental results show that the proposed method achieves consistently high accuracy for various distortion types. For simulated spectra with added noise, the mean center-wavelength error reaches 1.57 pm. For real sampled data, the average root mean square error (RMSE) of spectral fitting across multiple experimental groups is 0.006617. Compared with conventional peak-based methods, the proposed approach offers notable improvements in robustness, stability, and adaptability.

Conclusions This adaptive demodulation method provides an effective solution for handling distorted FBG spectra and significantly enhances the fault tolerance and operational reliability of FBG sensing systems. It also offers a promising pathway for extending sensor service life in complex environments.