Dynamic multilayer graph neural networks for fiber-Bragg grating dimension detection in aircraft applications sample

Fiber Bragg Gratings (FBG) dimension detections in an aircraft application involves for determining the precise structural parameters sensors to ensure the accurate monitoring of environmental conditions. However, existing methods face significant challenges due to the high sensitivity of FBGs to fabrication tolerances. Even minor deviations in structural dimensions can lead to substantial shifts in optical response characteristics, including quality factor, reflectivity, and sensitivity. These variations introduce inconsistencies in sensor performance, complicate the design process, and increase the risk of misclassification or undetected faults in critical aircraft applications. To overcome the issues, the FBG model is designed using ANSYS lumerical tool with Finite Element Method (FEM) analysis for determining the various dimensions based on the properties such as Quality factor, Sensitivity and Maximum Reflectivity which are evaluated and resulting data is collected for further analysis. Missing data due to simulation constraints are predicted utilizing Spatio Temporal Recurrent Neural Network (ST-RNN) to fill the incomplete data points to enhance the data quality. The imputed dataset undergoes Standard scaler Normalization to maintain nonlinear relationships. Next Transformer-based Time-Series Wasserstein Generative Adversarial Network with Gradient Penalty (TTS-WGAN-GP) performs data augmentation for generating the synthetic data that helps to increase dataset size for better model robustness. Coyote and Badger Optimization (CBO) used for feature selection which selects the key features to influence an optical response. Next DYMGNN model uses graph based neural network to classify the data relationship. The trained net model identifies fiber Bragg gratings based on input properties like quality factor, sensitivity and maximum reflectivity. Then output is predefined and classified as low, medium and high based on the clustered data for thickness, width, height and etching depth. The accuracy, Recall and FNR of the proposed method for synthetic and simulated data are 97.13% &96.98, 93.02% & 91.98, and 6.98% & 8.02, respectively. The proposed approach ensures an accurate classification of FBG designs for effectively modelling the complex relationship between structural parameters and optical response.