Skill & current project

Abstract: We are working on a novel liquid-level sensing system to enhance the capacity of the sensing system and reduce the cost. Our sensing system can monitor the liquid level of several points at the same time. Additionally, for cost efficiency, our system employs only one fiber Bragg grating (FBG) sensor in each spot. We use an FBG sensor which is connected to a properly designed float to detect liquid levels. When changing the liquid level inside the tank (water level container), the float position is changed, and the axial strain is applied to the FBG. The axial strain causes a shift in the reflected wavelength of the FBG sensors. Hence, the water level of the tank is monitored depending on the wavelength shift. The wavelength shift of FBG leads to overlap or cross-talk between two FBG water level sensors and is very challenging to properly identify the water level of each sensor. To solve this overlap problem and accurately predict each tank's liquid level, we proposed a Deep Neural Network (DNN) approach. The performance of the proposed DNN model is evaluated via different scenarios. The result proves that the proposed DNN model accurately predicts the liquid level of each spot.

Abstract: We work on a structure to absorb the THz wave and investigate its parameter by Deep Neural Network (DNN). The structure can absorb the incident wave and since there is a gold layer in the back with a thickness of d it is expected to the incident wave completely reflect, however, by adjusting design parameters it is possible to achieve an absorber structure. The structure is made by repeating a unit cell periodically in X and Y directions. The unit-cell structure consists of three bars of MoS2 with different intrinsic carrier Densities (N1, N2, and N3) and three different lengths (L1, L2, and L3). Artificial Intelligence (AI) method is used in two phases. In phase one we predict the reflection response of the structure in the forward path and in the second phase, we predict the design parameters for the desirable reflection inverse design path. In the forward path, the design parameters (N1, N2, N3, L1, L2, L3) feed to a deep neural network (DNN) named forward DNN with 4 hidden layers. We are preparing 5000 datasets for the training of the forward DNN which each data set consists of a design parameter as input and samples of reflection response as output.