Frequency Modulated Continuous Wave (FMCW) radar is a low power, compact mechanism which can be used for non-destructive health monitoring and inspection of surface and subsurface materials. This enables the detection of defects that are internal to the analyzed structural element and not visible. The key benefits of this technology are that it offers a non-contact monitoring tool at reduced costs, reduced risk and reduced time of inspection . Recent work has proposed to assess the capability of FMCW radar sensing for composite material characterization of wind turbine blades. While it showed promising results for the robust classification of a turbine blade of different thickness or inner composite materials, it was not yet applied in the context of health monitoring. In this work, we propose to study the feasibility of FMCW radar to detect anomalies in monolithic surfaces. This task utilized adapted signal processing and machine learning methods to analyze the return signal of the radar. In this work we propose a complex-valued autoencoder neural network with a new activation function adapted to the complex-valued input signal. The Autoencoder is trained on healthy samples only and the residual is used as a health indicator to distinguish healthy from surfaces with defects. To demonstrate the performance of our approach, we consider a monolithic composite containing engineered defects. We compare our anomaly detection strategies to other state of the art methods like support vector data description where our proposed approach demonstrates better performance. Moreover, we show that using the analytical representation of our return signal leads also to better performance than other signal return representations.