Distributed fiber optic sensors (DFOSs) based on Brillouin scattering have progressed significantly over the past decades and have seen many applications, particularly in the field of structural health monitoring. These applications often benefit from the capability to measure multiple parameters simultaneously, such as temperature, strain, and humidity. However, accurately measuring even a single parameter becomes challenging when the fiber is subjected to simultaneous changes in multiple parameters. This issue is known in the literature as cross-sensitivity. While solutions to mitigate cross-sensitivity in Brillouin DFOSs have been reported, they often increase the overall system’s cost or complexity by requiring the use of two optical fibers or specialty fibers. This thesis reports on the development of a machine learning-assisted Brillouin optical frequency domain analysis (BOFDA) system for simultaneous measurements of two or more parameters, including temperature, strain, and humidity. First, a BOFDA system capable of obtaining high signal-to-noise ratio (SNR) multipeak Brillouin gain spectra from a standard telecom optical fiber is presented. These spectra are used to extract features and train simple and robust machine learning models to simultaneously predict temperature and strain. Next, the ability of BOFDA to monitor humidity using a polyimide (PI)-coated optical fiber is demonstrated. This is the first application of BOFDA for humidity sensing and one of the few in general for DFOSs. The machine learning-assisted BOFDA is also shown to be effective in discriminating humidity and temperature in a single optical fiber using a similar methodology as for temperature and strain discrimination. Finally, by combining these methods, a solution for simultaneously measuring all three parameters (temperature, strain, and humidity) using standard acrylate-coated and PI-coated optical fibers is reported. All experiments were conducted in the lab under controlled temperature and relative humidity (RH) conditions, generally ranging from 20 ◦C to 60 ◦C and from 20% to 80%, respectively. However, the reported methodologies are not inherently limited to these ranges. Compared to time-domain systems, BOFDA is a cost-effective solution; however, its measurement time is significantly longer, often by up to two orders of magnitude. Although the main focus of the thesis is to address cross-sensitivity, the issue of long measurement time could hinder many applications and is therefore also considered here. It is shown that convolutional neural networks have the potential to significantly reduce the BOFDA measurement time, alleviating the major drawback of BOFDA. However, this solution has been developed only for single-parameter sensing. The development of a time-efficient multiparameter BOFDA sensor is seen as a promising direction for future research. The outcome of this research has significant potential for applications in structural health monitoring. Simultaneous multiparameter sensing with BOFDA could improve the safety, efficiency, and service life of critical infrastructure, such as bridges, tunnels, dams, highways, and submarine power cables.
Christos Karapanagiotis (Thu,) studied this question.