Distortion Tolerant Method for Fiber Bragg Grating Sensor Network Using Estimation of Distribution Algorithm and Convolutional Neural Network

In this paper, we proposed a distortion-tolerant method for FBG sensor networks based on Estimation of Distribution Algorithm (EDA) and Convolutional Neural Network (CNN). Addressing the parameter reconstruction of reflection spectrum, an objective function is formulated to pinpoint the Bragg wavelength detection problem, with the optimal solution acquired via EDA. By incorporating spectral distortion into the objective function, the EDA-based method effectively manages distorted spectrums, ensuring the fidelity of wavelength data. Further, CNN aids in extracting features from the entire FBG sensor network's wavelength information, facilitating the creation of the localization model. By sending the reliable wavelength data obtained by EDA to the trained model, swift identification of the load position is achieved. Testing revealed that under conditions of spectral distortion, EDA can adeptly detect the Bragg wavelength. Additionally, the CNN-trained localization model outperforms other machine-learning techniques. Notably, experimental results demonstrate that the proposed EDA surpasses the second-ranked method, i.e., the Maximum method, achieving a Root Mean Square Error (RMSE) of merely 1.4503mm which is substantially lower than the 6.2463mm achieved by the Maximum method. The average localization error remains under 2mm when 5 out of 9 FBGs' reflection spectra are distorted. Furthermore, Bragg wavelength detection error stays below 1pm amid spectral distortion. Consequently, our method offers promising application prospects for long-term FBG sensor network monitoring, ensuring high accuracy and robustness in detecting structural damage.