• Printed Journal
  • Indexed Journal
  • Peer Reviewed Journal
Journal of Applied Science & Engineering

Dhaka University Journal of Applied Science & Engineering

Issue: Vol. 6, No. 1, January 2021
Title: Discrete Wavelet Transform Based Artificial Neural Networks for Extracting Temperature Distributions in Brillouin Optical Time Domain Analysis Sensors

Keywords: Fiber optic sensors, Discrete wavelet transform, Artificial neural networks, Nonlinear least-squares fitting, Lorentzian profile

The uncertainty in extracting temperature distributions using Brillouin optical time domain analysis (BOTDA) sensors depends ultimately on the signal-to-noise ratio (SNR) of the BOTDA-measured Brillouin gain spectra (BGSs) along the fiber. The real-world applications of BOTDA sensors also require fast extraction of temperature distributions from the measured BGSs. To improve the SNR of the measured BGSs, 2D discrete wavelet transform (DWT) based wavelet denoising of BGSs (WDB) is used in this study. The denoised BGSs are then processed by using WDB-based artificial neural networks (WNNs) for the fast and accurate extraction of temperature distributions along a 38.2 km long fiber. The performances of WNNs are investigated in detail for the BGSs acquired from BOTDA experiment at ten different frequency steps and ten different numbers of trace averaging. The effect of WDB as well as WNN on the spatial resolution of the sensors is also analyzed. Moreover, the performances of using WNNs in extracting temperature distributions are compared with that of widely-used nonlinear least-squares fitting (NLF). The experimental results manifest that WNNs can offer much better uncertainty and significantly faster temperature extraction without sacrificing the spatial resolution as compared to NLF. Thus, the proposed WNNs can be effective tools for the fast and accurate extraction of temperature distributions in BOTDA sensors.

  1. ref1