DISTRIBUTED ACOUSTIC SENSOR USING CHIRP PULSE PHASE SENSITIVE OPTICAL TIME DOMAIN REFLECTOMETRY

Abstract

Fiber Optic Distributed Acoustic Sensor (DAS) is a strain and temperature sensor that can measure thousands of points simultaneously using conventional single-mode fiber as the measurement medium. This technology detects static and dynamic events such as time-varying signals and variations, various methods are available for collecting and analysing data from these distributed sensors. This thesis proposed a combined Image Processing and Machine Learning method to analyze and classify the strain data. This method exploits the Rayleigh backscattered power, a chirped pulses laser, and a -OTDR. In addition to that, an implementation of this distributed sensing system is also presented, which has been tested. The detected traces are successively analyzed to localize in the frequency, amplitude, and strain energy domain. The maximum acoustic signal frequency is between 2-125 Hz. Resonance signals are high-frequency signals used for testing and analyzing the data. Inside the fiber, 250 scattering points consider in one minute. Strain signals convert into the image to get a clear view. We use strain energy data present in the image and analyze the appearance of different intruders. Data is extracted in the form of all color images RGB (red, green, blue), HSI (hue saturation intensity), HSV (hue saturation value), and grey images. I also extracted Features of images like GLCM (grey label co occurrence matrix), LBP (local binary), Mean, Standard Deviation, Histogram Thresholding, and Canny filter at five different Thresholds, and Gabor filter. Different intruders are classified with the help of an SVM (support vector machine) classifier.