Intelligent design of intensity modulated fiber optic sensors using different fiber structures

Fiber optic sensors (FOSs) have opened a new era in sensor technology providing great opportunities and they have been widely used for many years. FOSs prefer many advantages over traditional sensors and due to these advantages they could be prefered instead of other sensors in several applications. FOSs can be divided into four groups according to their operating principle: intensity modulated, wavelength modulated, phase modulated, and polarization modulated fiber optic sensors. Phase, wavelength and polarization modulated FOSs are more sensitive, but they are expensive and require complex hardware and signal processing techniques. On the other hand, intensity modulated FOSs are simpler to construct and more economical. The major objective of this dissertation is to develop spatial and total intensity modulated FOSs with improved performance. Improvements of spatial and total intensity modulated FOSs using Image Processing Techniques, Artificial Neural Networks (ANNs) and different optical fiber structures have been investigated in this thesis. The emphasis is on the development of fiber optic statistical mode sensors. Several experiments have been conducted and analyses/computations have been performed during the works. Force and displacement have been measured during the experiments. Several force and displacement values have been applied to the sensors, minimum and maximum applied measurands change depending the experiments. Statistical mode (specklegram) sensors are among the sensors which are based on utilizing the spatial intensity change. Spatial information can be retrieved by the analysis of change of patterns distribution at the sensor output using image processing techniques with statistical features. Several researches have been conducted utilizing fiber optic statistical mode sensors. However, only two different statistical features have been used for speckle image analysis. Reported results generally were based on simple image differencing or correlation. In this thesis firstly, the use of various local statistical features has been proposed and local statistical features computed for various local regions of the speckle patterns in fiber optic statistical mode sensors. Secondly, new statistical global features have been proposed for the design of fiber optic statistical mode sensors. All statistical features have been computed by different image processing algorithms. The first and second order moments, which are pth order moments, have been proposed as statistical features in this thesis. These new features compared with correlation and image difference features which were reported in the literature before. The statistical features have been compared in terms of different characteristics: precision error, non-linearity, and hysteresis. Hetero-core fiber structures are alternatives to stripped or etched evanescent fiber structures and they are characterized by the simplicity in their fabricated structure. Hetero-core fiber optic sensors can be designed with increased sensitivity due to enhanced mode field coupling effect at hetero-core sensing section. Few researches have been conducted to exploit spatial content of hetero-core structure in fiber optic statistical mode sensors. In these works, only correlation statistical feature have been used and no statistical feature except correlation used in the sensor design. New statistical features have been proposed for the design of hetero-core fiber optic statistical mode sensors in this dissertation. Different hetero-core optical fiber structures have been produced and experiments have been conducted. Differencing, first and second order moment statistical features have been compared with correlation feature. Comparative analysis of using different hetero-core fibers in hetero-core statistical mode sensors and multi-mode fibers in statistical mode sensors has been performed. Altering the hetero-core section length of the hetero-core statistical mode sensors has also been investigated in the thesis first time in literature. Designing intelligent statistical mode/hetero-core statistical mode and intelligent bending sensors has also been fulfilled in this dissertation. In order to estimate the sensor response of these sensors, it is necessary to carry out lengthy and complex mathematical computations. To overcome these problems ANNs have been used as a computational paradigm to improve FOSs' performance and to develop intelligent FOSs, because ANNs can generate appropriate outputs for given inputs without any necessity to mathematical formulations between input and output data. With this technique distortions or deviations in the sensor system can be automatically detected, thus sensor becomes an intelligent system. Using data taken from the experiments conducted by the sensors and using ANNs, different intelligent sensor architectures have been proposed to predict the force and displacement values measured by the sensors. Intelligent sensor architectures with single input-single output and multiple inputs-single output have been utilized. Multi-layer Perceptron (MLP) with many algorithms, Radial Basis Function (RBF) and General Regression Neural Network (GRNN) ANN models have been utilized in the analyses. Performance comparisons demonstrate that all of the ANN models and algorithms utilized in this thesis could predict the measured values with considerable errors.