A Sensitized Plastic Optical Fiber Multi-point Bending Sensor Based on Deep Learning

Since the variations of mode interference induced by curvature in multi-mode fiber (MMF) can be well represented by the fiber specklegram. The paper proposes a fiber multi-point bending sensor based on deep learning to detect the bending of short-distance fiber and reduce the cost of detection. In the experiment, plastic fiber with sensitization processing was used as the transmission medium and CCD (Chargecoupled Device) is used to collect specklegram of different bending angles and different bending area at the end of the fiber output. A fiber of 60cm was bent at 15cm, 30cm and 45cm from the input, respectively. The convolutional neural network was used to classify the output speckle under different bending conditions. It was found after testing that the neural network can classify speckles with a bending interval of 15° with an accuracy rate of 94.3%. This method indicates the capability to distinguish the specklegram even when the fiber is under complicated bending due to the specklegram can represent the status of the whole section of fiber. The method proposed in this paper can be used in fields of short-range sensing or optical fiber multi-point bending measurement, it can also find applications in distinguishing the status of certain structures, such as robotic arms, mechanical finger and some disabled auxiliary equipment.