The Application Of Artificial Neural Network Modeling Of Conformal Array Elements

For today's high performance and fast electromagnetic engineering computer-aideddesign, it is increasingly necessary to model elements and devices with high accuracy,reliability and efficiency. Artificial neural network (ANN) recently has receivedextensive attention as a fast and accurate modeling tool in electromagnetic engineering.In this thesis, back-propagation (BP)-ANNs are uesd to model balun, coupler andso on. And knowledge-based neural network (KBNN) is introduced to modelsophisticated electromagnetic objects by incorporating prior knowledge into artificialneural structures. Also a KBNN model is constructed for the Schiffman phase shifter.Firstly, the BP neural network model is used for modeling the balun, coupler, ultrawideband antenna, and reconfigurable antenna. Results show that the network trainingprocedure becomes efficient and stable and the developed ANN models are robust forgeneralization.Secondly, the NNKBN trained with insufficient training data is applied to modelthe Schiffman phase shifter. Results show that the NNKBN models are good forextrapolation with high accuracy.Finally, we make some changes according to the original Schiffman phase shifter'sshortcomings and design a modified one. This novel structure realizes a 90°phaseshift with a low permittivity substrate and also can make the fabrication easer for bothcases of low and high permittivity substrates.All of the proposed models not only preserve the accuracy of the EM simulations,but also reduce their CPU and memory requirements, and at the same time keep goodextrapolation capability. Hence, the artificial neural networks of the problems to bemodeled have potential power for the high performance and fast CAD inelectromagnetic engineering.