A Deep Learning-Enhanced Multimode Fiber Bragg Grating (MMFBG) Bidirectional Bending Sensor

In this article, a deep learning-enhanced multimode fiber Bragg grating (MMFBG) bending sensor is proposed with the capability of bidirectional bending sensing. Unlike fiber Bragg gratings (FBGs) in single-mode fibers (SMFs), MMFBGs exhibit multiple reflected peaks in their reflected spectra, especially when core-offset launching is employed. When various bending is applied to the grating, the couplings between the forward and backward fundamental modes and higher order modes are affected, which results in a changed reflected spectrum. Therefore, the reflected spectrum with rich information can be used to characterize the bending. The MMFBG is encapsulated using a silicone tube filled with polydimethylsiloxane (PDMS) to protect the grating and enhance the sensor’s stability. In order to track the nonlinear relationship between the spectrum and the bending of MMFBG, a deep learning model is utilized integrating a 1-D convolutional neural network (1D-CNN), a bidirectional long short-term memory (BiLSTM) network, and a self-attention mechanism, which enables accurate prediction of bidirectional bending angles. The experimental results demonstrate that within the bending range from −135° to 135°, the proposed model achieves a coefficient of determination ( ${R} ^{{2}}$ ) of 0.9999 and a root mean square error (RMSE) of 0.8022° in bending prediction. For unknown bending angles, the RMSE is 4.0094°. Compared with traditional wavelength monitoring and deep learning based methods, the proposed sensing scheme enables a more efficient and accurate sensing of bidirectional bending with a very simple structure.