Geometry Verification of Arbitrary Shaped Conductive Cavities by Means of Eigenmodes classified with Artificial Neuronal Networks

In this paper a method for fault geometry verification and fault detection based on geometric feature extraction and neural network classification is proposed. The geometric feature extraction is done by means of broadband s-parameter measurements followed by eigenmode extraction for geometry verification of arbitrary shaped conductive cavities. The target is a classifier that is fault detection of fabricated arbitrary shaped conductive cavities with the help of a simulation based database only. For a proof of concept a simplified geometry model with four variable dimensions was defined. With the help of CST Studio Suite a large database consisting of 625 different simulations was created. In a second step, the simulation data is reduced to eigenmodes that are further investigated with respect to the frequency change per geometry deviation. A classifier has been developed from the simulated eigenmodes of the scattering transmission parameter |S21| with the aid of a multilayer feed forward neuronal network. Considering five selected eigenmodes, the classifier can detect the geometry parameters of six reference channels with absolute geometric deviations of 1 mm or less.