Fiber optic sensors and analysis of sensor parameters with artificial neural network based optimization algorithm

Fiber optic sensors are vital for detecting surrounding changes in temperature, strain, vibration, chemical, gas, refractive index, and so on. Surface plasmon resonance (SPR) sensors, a form of fiber optic sensor, are used in very sensitive applications such as biological, chemical, and refractive index change detection. It is essential to optimize the sensor parameters that influence efficiency and sensitivity. In the literature, the optimization of SPR sensor parameters is done by the conventional method, which is based on scanning one parameter at a time and keeping the rest constant and applying this technique one by one for each parameter. In this method, the correlation between parameters cannot be observed and the result may not be the optimum because one parameter is scanned at a time. Therefore, the performance of the sensor may not reach the desired high sensitivity. In the last decade, artificial intelligence-based optimization approaches have become very popular and it would be wise to use a hybrid artificial neural network-genetic algorithm (ANN–GA) to solve optimization problems that may occur. In this thesis, a python code for a hybrid ANN-GA structure was generated. With the conventional optimization approach, the best values of N, d, Ag_th, and a parameter were observed as 20, 50 nm, 70 nm, and 10 nm at refractive index values of 1.35 and 1.39, respectively, and the sensor sensitivity was obtained as 3775 nm/RIU. In this thesis, the data set obtained by FEM with the hybrid ANN-GA based optimization approach we proposed was trained and tested with R2 values of 97.83% and 96.84%, respectively. In this developed model, N, d, Ag_th, and a parameter were estimated as 20, 50 nm, 75 nm, and 10 nm, respectively, and as a result, 3890 nm/RIU sensitivity was reached.