Intelligent Microwave Detection of Surface and Sub-Surface Anomalies

Abstract

Microwave near-field testing is a promising nondestructive testing method because of its unique capability to interrogate metallic surfaces and multi-layer dielectric structures. Due to today's need for lighter, stronger, and more-durable materials, enhanced dielectrics are increasingly being used to replace or coat metals. Consequently, conventional testing methods, with their limited penetration, are no longer adequate, but microwave testing sensors transmit signals that can penetrate into dielectrics and so detect surface and subsurface anomalies.

Due to the growing use of microwave near-field sensors for different daily life applications, there is an ongoing need to improve their performance. Recently, artificial engineered electromagnetic materials (metamaterials) have been utilized to demonstrate strong localization and enhancement of electrical fields around sensing elements in order to improve probes sensitivities. Metamaterials are being used to enhance sensors design at the hardware level for better anomaly and flaw detection. Currently, microwave sensors are being used to capture large and complex information, but doing so requires better integration of signal processing methods. Implementing artificial intelligence algorithms to process information collected by microwave sensors can address the challenge associated with information complexity or obscure pattern changes.

To address this gap in microwave near-field evaluation, this study integrates machine learning techniques with microwave near-field testing. Machine learning is a subset of artificial intelligence that denotes a set of methods that can automatically detect patterns in data to build a learning model. The learned model is then used for decision making about unseen data. Employing machine learning techniques for building classification models, this work combines machine learning algorithms with microwave near-field testing. In particular, it aims to build machine learning models that enhance flaw and anomaly detection in microwave near-field testing. The trained machine models can be integrated or embedded in a portable device or rack mounted microwave near-field testing equipment. The value of this approach is confirmed through numerical simulations and laboratory measurements.