Design and Performance Analysis of SMS-Based Fiber Optic Sensor for Small-Scale Bending Detection Using Machine Learning

This study presents the design and performance analysis of a fiber optic bending sensor based on Single-mode–Multimode–Singlemode (SMS) structure, integrated with a machine learning algorithm for bending detection based on its curvature level. The sensor is constructed by splicing a multimode fiber (MMF) between two singlemode fibers (SMFs), forming a multimode interference pattern (MMI) that is sensitive to geometric deformation. The experimental setup measures spectral shift and other features such as curvature, angle, and sensitivity. Random Forest (RF) and Support Vector Machine (SVM) classifiers are applied to predict the bending level, with RF achieving 100% accuracy and SVM achieving 99% after SMOTE balancing and hyperparameter tuning. Feature importance and SHAP analysis confirm the dominant role of geometric features over optical features in predicting the curvature class. Although the spectral shift is very small. The integration of the SMS sensor with a machine learning model is proven to be very effective for bending detection based on the curvature level at a small scale.