Profiling Wiper Arm Surface for Appearance Defect Detection

Defect detection in the wiper arm manufacturing industry is not highly automated due to the task complexity. In addition to its non-uniform structure, a wiper arm is usually dark and has a reflective surface. Prior work on area scan camera showed images with narrow visible zones and small regions of interests (ROI). Consequently, a lot of sequential images need to be taken consecutively and processed. Despite many effort to improve its performance using various image processing techniques, the defects that occur at the boundary of the ROI were not detected as they blended into the background during binarization. This paper proposed to use a line scan camera to capture images that resulted in a much wider visible zone. The total number of images to be captured per wiper arm has significantly reduced from 66 to 10. The image processing time per wiper arm has also improved tremendously from 7.45 seconds to 0.067 seconds. With considerable decrease in the miss-detection of this type of defects, the overall accuracy has improved from 0.905 to 0.935. Meanwhile, the classification accuracy for scratch, bump and dent is 86.16%, 97.06% and 98.02% respectively. This shows that a line scan camera performs better in this application, even though the set-up is more complicated.