A robust machine vision system design to facilitate the automation of surface appearance inspections

The performance demands on coated surfaces (e.g., automotive paints) are considerable. Though the primary purpose of automotive paints is to protect the car body from corrosion, it has been well established that the appearance of a painted surface greatly affects a customer's perception of that product's quality. Automotive manufacturers spend considerable sums on the painting process during manufacturing and, again, on warranty-covered claims of paint-related problems. Therefore, robust methods to both characterize and monitor surface quality are critical. Existing quality control methods expert inspectors are generally effective in assessing perceived appearance; however, they are often labor intensive, time-consuming and can be subjective. Optical techniques conducted offline are generally robust and objective; however, they are not always effective in assessing perceived surface appearance. Furthermore, neither option effectively determines and maintains optimal processing conditions to consistently produce a quality finish and mitigate defects. An intelligent automated process that incorporates the effects of perception can facilitate an effective analysis of surface quality and appearance, in addition to maintaining optimal processing conditions for producing a desirable appearance. A long-term research goal is the development of a robust, automated, in-line monitoring system which controls critical painting process parameters based, in part, upon captured image data which correlates strongly with human visual assessment. As a significant first step, this paper presents captured images and derived image attributes that correlate strongly with both objective measurements and a visual ranking of specular painted samples.