Defect detection and classifcation of zinc coated parts using machine vision

A machine vision approach is required for detection and classification of surface defects such as pitting, watermarks, rough surface and color tonality in zinc coated parts. Such an automated visual inspection scheme involves following steps: image acquisition, defect segmentation, feature extraction and classification. Image acquisition of a part with complex geometry and specular surface depends on the effectiveness of illumination. Various illumination techniques such as dark and diffuse with various incident angles were tested for capturing images with maximum contrast and information. Feature extraction of the defects was then carried out to capture the brightness, shape and spectral properties of the defects. Then feature ranking was calculated to obtain the best features for a real-time inspection. Supervised classification was then tested with Support Vector Machine (SVM) for defects other than tonality difference. The color tone is unique to an inspection batch and hence a change in tonality had to be detected in-line. An unsupervised classification method was used as an extra step with in-line learning for this case. Challenges like complex geometry, specular reflection and real-time processing in building the machine vision system for real time inspection is solved by developing various image-processing algorithms. The color tonality inspection is a unique problem faced in this application and that was addressed at two stages by both supervised and unsupervised classification techniques.