Detecting Failed Elements in an Arbitrary Antenna Array using Machine Learning

Abstract

The failure of antenna array elements leads to inaccurate far-field radiation patterns. When failed elements are present in an antenna array used for sensitive applications, it will compromise the measured data, which may severely affect the end-result. Identifying faulty elements is therefore important for the system health management of the antenna array. In [2], Feedforward Neural Networks (FNNs) were used to classify failure scenarios in an antenna array, using the far-field radiation pattern. Different methods of sampling the far-field pattern were compared according to the resultant accuracy. In this article, Support Vector Machines (SVMs) with low degree polynomial kernels, as described in [1], are trained on datasets generated by different sampling methods. The work is tested on arbitrary array configuration simulations with an increasing number of elements, to investigate the possibility of using the SVM method described in [1] on large antenna arrays.