An Efficient Model for Inspecting the Quality of Pellet Components Automatically

抄録

This paper proposes an efficient method based on neural networks to automatically determine the quality of pellet components. By using two images (front and side views, referred to below as CIFV and CISV respectively) captured through the microscope using a CCD camera, it is possible to assess the quality of pellet component. Here we define the captured image in front view as CIFV and the captured image in side view as CISV. In addition, Two corresponding template images are defined here as TIFV for front view and TISV for side view, respectively. These template images were created based on the design dimension. Prior to carrying out the quality evaluation, the captured images were processed to eliminate noise and the images were binarized. Then, the datum matching plane and the datum machining line for the captured images were calculated. Next, using a coordinate transformation related to the positions of the datum plane and the datum line, the test hole in the captured image can be moved to the center part of this image in order to match the template image. After that, the values of certain feature parameters are calculated based on the errors and on the differences between the captured images and template images. Finally, the values of parameter features serve as inputs of the neural networks for determining the quality of the component. The experimental results confirm the effectiveness and efficiency of this procedure for inspecting the quality of pellet components.