Sensitivity Identification of Low-Frequency Cantilever Fibre Bragg Grating Accelerometer using Cascade-Forward Backpropagation Neural Network

Abstract

The frequency-dependent issues and instrumentation requirement for FBG sensors necessitate the identification of the sensitivity of the cantilever FBG accelerometer using machine learning. As result, this article presents a cascade-forward backpropagation (CFB) neural network with an orthogonally-phase chirp signal with a range of constant forcing frequency and steadily increasing base acceleration amplitude as its input. This input/output data set was numerically calculated by integrating modal model and Euler-Bernoulli beam approach (FBG-MM). The maximum amplitude of the base acceleration was 200 m/s² and the forcing frequencies and location of the FBG sensor mounted on the beam measured from the fixed end were 1 to 90 Hz and 0.03 m, respectively. The trained CFB predicted the wavelength shift very well, but it was restricted to one-half of the forcing frequencies of those used in the CFB training process, whereas the base acceleration is not an important element in determining the sensitivity of the FBG accelerometer. In terms of the FBG sensor’s location on the beam, considering a few positions will greatly expand the CFB’s capabilities. Future work will include the use of the trained CFB as  “black-box sensitivity” for actual acceleration measurement, as well as the use of empirical data to replace the numerical FBG-MM as the input/output training data set.