Study on Visual Evaluation for Barrel Finished Surfaces

Abstract

In this paper, an intelligent visual evaluation system using a neural network is developed in order to evaluate the barrel finished surfaces. It is well known that the surface integrity after barrel finishing depends on type of finishing machines, nature of the media and the compounds. The integrity of barrel finished surfaces, when different compounds were used, have typically been evaluated by visual inspection of expert workers. It is necessary to introduce the optical qualitative evaluation system instead of visual inspection in an automated inspection line. However, it is very difficult to evaluate the finished surface using image processing and/or other measuring equipment, because of complexity of data analyses for slightly different light reflection data among each finishing condition. As a result, not many papers have been published on the optical evaluation system for the barrel finished surfaces when different compounds were used. The objectives of the paper are to analyze the image of barrel finished surfaces using image processing, and to construct a neural network system which is able to evaluate the finished surface integrity. In this paper, it is found that features of barrel finished surfaces are decided by image analyses. As a result, it is possible to construct a neural network system composed of the finished surface image features as input layer and grades by the sensory test as output layer.