Nanoparticle-doped fibers combined with machine learning methods for 3-dimensional shape reconstruction systems

The development of nanoparticle (NP) doped optical fibers provides new possibilities in distributed sensing and provides important advantages in spatially distributed system due to the tailoring of the backscattering signal, which enables the development of new approaches in shape reconstruction systems. For this reason, this paper presents the application of a distributed sensor system based on optical frequency domain reflectometry (OFDR) using the NP-doped optical fiber as the sensing medium in which there is an increase in sensitivity and spatial resolution due to the increase in backscaterred signal. The fiber is embedded in a rubber strip to increase the measurement range and the sensor robustness due to the nitrile rubber flexibility, since the rubber can provide strain transfer to the optical fiber, leading to higher strain limits, which can withstand the complex 3D loadings for the shape reconstruction. The sensor system is tested under different mechanical loadings conditions and the contribution of this work is a readily available sensing system using specialty optical fiber with high spatial resolution for 3D shape sensing approaches. The results indicate the feasibility of the proposed approach for shape sensing monitoring through a determination coefficient ( R 2 ) higher than 0.99 for all cases. In addition, the strain distribution along the fiber was also estimated for all loading conditions as well as the combination of them with an RMSE of around 1 mm considering the amplitude estimation of all loadings. The 3D shape reconstruction resulted in a mean error of around 2.0% considering all 3 axes of the cartesian plane. Therefore, the proposed sensor system is a feasible option for shape sensing with sub-centimeter spatial resolution and high accuracy using the NP-doped optical fiber as enhanced backscattered medium. • Development of OFDR system using single NP-doped optical fiber as sensing medium. • Machine learning implementation for 3D shape reconstruction with a single fiber cable. • Rubber embedment to increase the measurement range and sensitivity of 3D shape sensing. • 3D shape reconstruction with mean error of around 2% considering all planes.