A Mach-Zehnder-based optical fiber sensor enables multi-directional and wide-range curvature sensing with the assistance of CNN

Abstract

In this work, wide-range curvature sensing in multiple directions was achieved by combining a Mach-Zehnder-based optical fiber sensor (OFS) with convolutional neural networks (CNNs). The experimental results show that the curvature sensitivity of the proposed sensor is − 8.8320 nm/m−1 in the range of 0–3.1171 m−1. The application of CNNs significantly improves the measurement range to 0–10.4167 m−1, and the sensor can realize measurements over a wide curvature range in four directions. The accuracy of the measurements can still be guaranteed despite a sudden decrease in the number of wavelength sampling points, and curvature direction recognition was realized. Further, OFSs can achieve a wide range of curvature sensing in 12 directions by taking full advantage of the information contained in the transmission spectrum. The proposed technique is highly versatile and conducive to the popularization of OFSs in the field of curvature measurement.