Machine Learning Applied to Terahertz Signal Analysis For Hygrothermal Aging Characterization of Composites

Abstract

Given their noncontact nature, ease of inspecting nonmetallic materials, and ability to detect small structural changes, terahertz waves were chosen as the study methodology for this project. Initially, characterization data was pretreated and standardized. The data were then reduced in dimensionality using principal component analysis and classified using other machine learning methods. The results demonstrated that the terahertz wave technique could accurately distinguish between different temperature and aging time scenarios. They also indicated that nondestructive in-service characterization of composite repairs is feasible and can provide invaluable information for decision-making. The main benefits of this approach include ensuring the safe operation of offshore piping and optimizing resources when deciding whether to replace the repair or keep it in service.