A Method for Improving Performance of Vector Bending Fiber Sensor Based on LSTM

This study focuses on achieving accurate and reliable prediction of the results from vector bending sensors by utilizing a neural network (NN) model to precisely forecast sensor curvature over a wide dynamic range. The analysis of the sensor’s spectral response is conducted through the application of NNs. A total of 576 datasets were considered, incorporating 2501 input parameters, and these datasets were expanded using the Bootstrap resampling method. The sensor data of varying lengths and bending angles were predicted using the long short-term memory (LSTM) model. The investigation also covered the impact of input time-series length on model performance, aiming to reduce the hardware requirements for the sensor units. Model assessment was carried out using the absolute coefficient (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${R}^{{2}}$ </tex-math></inline-formula>), mean absolute error (MAE), and mean bias error (MBE) as evaluation metrics. After training the model on multiple datasets, the results indicate that processing spectral data with the LSTM network increases the likelihood of developing faster, cheaper, and simpler curvature detectors using coarse-resolution, high-speed spectrometers.