Browse Prior Art Database

Robust Neural Printed Circuit Board Analysis

IP.com Disclosure Number: IPCOM000101120D
Original Publication Date: 1990-Jun-01
Included in the Prior Art Database: 2005-Mar-16
Document File: 2 page(s) / 45K

Publishing Venue

IBM

Related People

Kishi, GT: AUTHOR [+2]

Abstract

An inspection method is described that uses a neural network to analyze printed circuit boards. The neural network is used to analyze a digitized video image of a via hole to determine whether or not a component pin is present. This is used to determine if the component was successfully inserted.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 73% of the total text.

Robust Neural Printed Circuit Board Analysis

       An inspection method is described that uses a neural
network to analyze printed circuit boards.  The neural network is
used to analyze a digitized video image of a via hole to determine
whether or not a component pin is present.  This is used to determine
if the component was successfully inserted.

      A back-propagation neural network was used to process the
image.  It consists of three layers:
     input layer
     This layer was mapped directly to the digitized pixels
     for the image.
     middle layer
     output layer
      This layer was mapped into the following categories of
     images:
      -  Empty square pads
      -  Empty round pads
      -  Pin Grid Array pins
      -  Single Inline Package (SIP) pins
      -  Dual Inline Package pins clinched upwards
      -  Dual Inline Package pins clinched downwards
      -  Dual Inline Package pins clinched right
      -  Dual Inline Package pins clinched left

      Examples of each of the different categories of images were
taught to the back-propagation network.  When the training was
complete, the network was able to distinguish between each of the
trained types of pins and empty holes.

      Once each of the specific images is learned, the neural network
is used to analyze the image as normal, but the pass/fail decision is
not made by comparing the expected (e....