Using Encoder-Decoder Convolutional Networks to Segment Carbon Fiber CT

Daniel Sammons, William P. Winfree

Feb 18, 2016 (modified: Feb 18, 2016) ICLR 2016 workshop submission readers: everyone
  • Abstract: Materials that exhibit high strength-to-weight ratio, a desirable property for aerospace applications, often present unique inspection challenges. Nondestructive evaluation (NDE) addresses these challenges by utilizing methods, such as x-ray computed tomography (CT), that can capture the internal structure of a material without causing changes to the material. Analyzing the data captured by these methods requires a significant amount of expertise and is costly. Since the data captured by NDE techniques often is structured as images, deep learning can be used to automate initial analysis. This work looks to automate part of this initial analysis by applying the efficient encoder-decoder convolutional network at multiple scales to perform identification and segmentation of defects for NDE.
  • Conflicts: