| Abstract Scope |
Composites are essential for developing space structures due to the amount of mass that can be preserved while sustaining structural integrity. As these materials have a complex layout for optimal performance, the life-cycles of these vehicles depend on the integrity that a composite assembly can endure. When damage is visible on the surface, visually identifying if the vehicle is safe to fly is a simple task; however, Non-Destructive Evaluation (NDE) is required to evaluate potential damage underneath, even when the vehicle seems nominal on the outside. Through Ultrasonic Transducer (UT) inspection, a pristine sample of a thin composite plate and a defected sample can be observed as they differ when waves are propagated through the body; machine learning (ML) can then be applied to quantify damages from different defect studies to detect potential patterns. In this study, different methodologies of ML including regressions and classifications were used to approximate a defect’s size, type (i.e. delamination versus disbond), and location from the UT source based on a library of varying defect simulations quantified into multiple dimensionless quantities when comparing waveforms: Damage Index in the field of NDE, or Featured Extraction in ML. When the dimensionless values are quantified as a relationship between the pristine and damaged cases, material properties and geometry will no longer be significant, guaranteeing a Convolutional Neural Network that can be expanded and applied for any thin plate inspection, given that it is through Lamb Waves, and applied for forecasting a defected region. |