A New Method for Fast Spectral Demodulation of Wide-Measurement Range Optical Fiber Torsion Sensor

Abstract

We present a new method using the combination of the fast Fourier transform (FFT)-support vector regression support vector regression (SVR) algorithm for fast spectral demodulation of an optical fiber torsion sensor based on Sagnac interferometer (SI). Experimental results demonstrate that with the aid of the FFT-SVR algorithm, the full torsion angle range from −360° to 360° can be predicted with a mean absolute error (MAE) of 3.05° and determination coefficient of 0.9995. More importantly, compared with the SVR algorithm-only demodulation, the FFT-SVR algorithm can efficiently decrease the number of features from 2001 to 25, and thus, the running time can be decreased from 0.005 to 0.002 s, which makes the running time a 60% reduction. Additionally, the spectrum scanning time can be decreased from 1.6 to 1 s, thereby resulting in a 37.5% decrease in the spectrum scanning time. Compared with other FFT-based machine learning methods, the FFT-SVR algorithm has high measurement accuracy and short running time, which greatly improves the measuring speed of optical fiber torsion and has great potential for application in many fields, such as the attitude sensing, shape measurement, and automatic control of robotic manipulators.