Quantification of Low Velocity Impact (LVI) Delamination from Ultrasound (UT) Inspection Data and Machine Learning

Abstract

In this paper we demonstrate the use of machine learning for segmentation of raw ultrasound scan of impacted composite panels. The methodology implemented describes the use of data augmentation, where sparse real data is augmented with synthetic data, to enable use of convolutional neural network for machine learning. Initial results indicate that the approach is able to segment the delamination patter with good accuracy.