Artificial-Neural-Network-Assisted Distributed Directional Optical Fiber Torsion Sensor With the SSAF-Based Sagnac Interferometer

A distributed optical fiber torsion sensor assisted by an artificial neural network (ANN) is designed based on the single stress-applying fiber (SSAF)-based Sagnac interferometer and demonstrated experimentally. A theoretical model is established to describe the piecewise torsion of the high birefringent fiber in the Sagnac loop, and the ANN algorithm is employed to demodulate the distributed torsion signals. The experimental results demonstrate the ability to simultaneously measure the torsion position and torsion angle when only one position along the fiber is subjected to torsion. The average R
²
and MAE of three predictions of the single torsion position are 0.984 and 0.55 cm, and those of the torsion angle are 0.998 and 1.98° for torsion angles from -120° to +120°. When two positions along the fiber are subjected to torsion, the R
²
and MAE of the torsion angle are 0.96/11.31° and 0.99/2.17° for torsion angles between -120° and +120°. The sensing system is demonstrated to demodulate the torsion of the wrist joint and elbow joint of the human arm. This promising strategy not only achieves the highest torsion position resolution and shortest response time, but also can be used to perform distributed torsion measurement and torsion direction recognition at the same time in addition to measuring the torsion at multiple positions, thereby showing great potential in applications such as robotic arms.