Surface defect identification and measurement for metal castings by vision system Academic Article in Scopus uri icon

abstract

  • © 2017 Society of Manufacturing Engineers (SME) An inspection system based on vision technology was developed to identify defects on the surface of a metal part produced by a casting process. In the proposed methodology, binary images of the bright and dark regions of the surface are first obtained. Connected components of these images are processed to find the shadows originated from defects. The algorithm to process the binary images was implemented on a Jetson TK1 board, and programmed in CUDA. The setup performs the computation in 900 ms for images of 5 megapixels, and the connected components algorithm is three times faster compared to commercial software running on a CPU. The parameters to find the shadows are independent of the field of view and resolution, i.e., the quantities that relate the two binary images can be expressed in pixels.

publication date

  • January 1, 2018