NDE inspection of AFP manufactured cylinders using an intelligent segmentation algorithm

Abstract

This article will discuss the approach and results for the identification of defects on an Automated Fiber Placement (AFP) manufactured cylinder. While the increase in productivity in AFP manufactured structures has allowed for the large-scale production of composite parts, imprecision in process can lead to the production of defects. A comprehensive platform for the inspection of AFP manufactured cylinders for utilizing a profilometry-based data collection approach was used. This data is then processed by a novel machine learning method. The machine learning method is based on the creation of fully convolutional neural networks and is used to fully characterize defects developed on cylindrical parts. Defect information was used to inform the repair of the cylinders and capture defects on cylinders that were intended to have hand placed defects. Cylinders were then used for validation of structural analysis tools.