Improving surface detection for quality assessment of car body panels

Surface quality analysis of exterior car body panels was still characterized
by manual detection of local form deviations and subjective evaluation
by experts. The approach presented in this paper is based on 3-D image
processing. A major step towards automated quality control of produced
panels is the classification of the different kinds of surface form deviations.
In previous studies we compared the performance of different soft computing
techniques for the detection of surface defect types. Although the dataset
was rather small, high dimensional and unbalanced, we achieved promising
results with regard to classification accuracies and interpretability of rule
bases. In this paper we reconsider the collection of training examples and
their assignment to defect types by the quality experts. For improving the
reliability of the defect classification we try to minimize the uncertainty of the
quality experts’ subjective and error-prone labelling. We build refined and
more accurate classification models on the basis of a preprocessed training
set that is more consistent. Improvements in classification accuracy using a
partially supervised learning strategy were achieved.