1. Dynamic strain measurement using Brillouin optical time-domain reflectometry
- Author
-
Li, Bo, Chu, Daping, and Soga, Kenichi
- Subjects
fibre optic sensing ,Brillouin scattering - Abstract
Due to the wide range of applications in structural health monitoring and the advantages of stability and flexibility in harsh environment, the Brillouin scattering based distributed fibre optic sensors have attracted significant attentions in recent decades. These sensors can quantitatively measure the distributed strain and temperature information along the optic fibre. The Brillouin optical time-domain reflectometry (BOTDR) has the advantages of single-end access and simple operation at the construction site, which makes it very popular in civil engineering applications. Furthermore, the demand of the dynamic sensing of strain information is becoming stronger and stronger, along with the rapid developments in some civil engineering fields such as geophysical sciences and the oil and gas industries. Therefore, the dynamic measurement of strain using the BOTDR system is a very practical and promising topic, especially for the civil engineering related industries. In this study, a BOTDR optical system is set up. A high extinction ratio modulator is chosen to overcome the drawbacks of the power leakage caused by the modulator. A high-speed polarization scrambler is used for the stable dynamic measurement of the BOTDR system. The coherent detection is adopted in this setup to boost the scattered Brillouin power from the optical fibre. A method realizing the distributed dynamic strain measurement of BOTDR is proposed and experimentally demonstrated using small gain stimulated Brillouin scattering (SBS) based on the short-time Fourier transform signal processing algorithm. The concept of small gain SBS is put forward for the first time. The backscattered Brillouin power from the fibre under test and the signal-to-noise ratio of the system output are enhanced by the small gain SBS. The input power limits of the BOTDR are discussed, and it is found that the modulation instability is the dominant input threshold for this BOTDR system. The frequency uncertainties of the Brillouin frequency shift, with given pulse durations, fibre lengths and number of averaging, are calculated. The 60Hz sinusoidal strain vibrations on the optic fibre under test are experimentally detected with 4m spatial iv Word Template by Friedman & Morgan 2014 resolution and 5.1MHz Brillouin frequency uncertainty. The scattered Brillouin power is captured with different input optical power, indicating the induced small gain SBS on the optic fibre. Then, an image denoising method using the convolutional neural network (CNN) is applied to the derived Brillouin gain spectrum (BGS) images to enhance the performance of the Brillouin frequency shift detection and the strain vibration measurement of the small gain SBS BOTDR system. By reducing the noise of the BGS images along the length of the fibre under test with different network depths and epoch numbers, smaller frequency uncertainties are obtained, and the sine-fitting R-squared values of the detected strain vibration profiles are also higher. To further improve the detection of the vibration, the BGS images along the time axis are drawn for the fibre section with vibration and denoised by the neural network. With the context of neighbouring spectra over time during the denoising process, the detected strain vibration profiles of the BOTDR system are smoother with the smallest Brillouin frequency uncertainty of 2.32MHz and the sine fitting R-squared values are enhanced with the maximal value of 0.931. After that, the methods to expand the receptive fields of the denoising CNN without deepening the network are introduced to find a way to further denoise the BGS images and boost the strain vibration detection of the BOTDR system. The two main methods of the U-net CNN denoising and the dilated CNN denoising are used for the BGS image denoising respectively. The U-net CNN demonstrates a worse detection of the Brillouin frequency shift, as the local information is lost during the denoising process. On the contrary, the dilated CNN method leads to even better detected frequency accuracies and better sine-fitting r-square values of vibration profiles, compared with the denoising CNN method. By denoising the BGS images along the time axis over the fibre section with vibration, the detected strain vibration profiles are more sinusoidal with the maximal R-squared value of 0.966, and the frequency uncertainty is further improved to 2.21MHz.
- Published
- 2021
- Full Text
- View/download PDF