Machine-Learning Based Estimation of the Bending Magnitude Sensed by a Fiber Optic Device

Bending estimation is an important property that must be assessed in several engineering applications including structural health monitoring, aerospace, robotics, geophysics, etc. While strain gauges and accelerometers are used to estimate bending behavior based on Machine-Learning (ML), few works in the literature have focused on the estimation of the magnitude of bending by combining ML techniques and fiber optic sensors. In this work, an ML-based method for estimating bending magnitude using the signal generated by an optical fiber sensor is presented. The sensor is formed by splicing a single-mode fiber with a multimode fiber. The interferogram generated from the sensor is processed to create a set of signal feature vectors (FVs). Thus, for estimating the bending magnitude, these FVs are used to train Machine-learning algorithms including Support Vector Machine, K-Nearest Neighbors, Naive Bayes, and Random Forest. To evaluate how each ML model performs, the accuracy, precision, recall, and F1-score metrics are used. The best performance is obtained by the Random Forest algorithm with a classification accuracy of 100%.