Sensitivity analysis of designed apodized fiber Bragg grating sensor using artificial neural network and tree-based models

An apodized FBG is designed to eliminate the side lobes from the reflectivity spectrum for the sensitivity measurement of physical parameters including temperature, strain, and pressure. The impact of the proposed apodization function is discussed with several other apodization profiles for the elimination of side lobes from the FBG spectrum. The sensitivity of the measurand is measured by analyzing the changes that occur in the Bragg wavelength, which is sensitive to the physical parameters. A strong linear response has been observed for the sensitivity of temperature, strain, and pressure for the designed apodized FBG. Machine Learning (ML) methods such as Artificial Neural Network (ANN) and Tree-based models are implemented for the predictive analysis of physical parameters and their respective Bragg wavelength shifts for the reliability of the designed apodized FBG sensor to achieve improved sensing outcomes, particularly to work in a hazardous environment for understanding any adverse scenario of a measurand. Decision Tree (DT) and Random Forest (RF) are used as Tree-based models in this work. Different statistical measures are introduced to evaluate the predictive analysis performance of the ML models. It has been observed that the predictive analysis performance of the ANN is better compared to the Tree-based models.