Curvature sensing via symmetric fiber ball MZI and neural network with sparse data

Fiber optic curvature sensors face several challenges that limit their application in structural health monitoring. These include high fabrication and demodulation costs, poor stability and robustness. In this paper, we propose a novel Mach–Zehnder interferometer (MZI) sensor that overcomes these limitations.The sensor was fabricated through the splicing of a seven-core fiber(SCF) and two single-mode fibers (SMF) with a common fusion splicer. The proposed sensor exhibited a good linear response, with a sensitivity of 22 nm/m−1 in the range of 0–0.313 m−1. The MZI sensor was demodulated by a neural network-based system and a data augmentation algorithm, reducing the need for large data samples and improving demodulation accuracy on sparse datasets. Overall, the sensor is characterized by fabrication simplicity, high mechanical strength, and compatibility with existing optical devices, making it applicable to various fields, such as aerospace, civil engineering, and mechanical manufacturing.