An adaptive clustering method detecting the surface defects on linear guide rails

Abstract

ABSTRACT Linear guides are widely used in automation, aerospace production, and medical care. It is currently one of the most concerning issues how to monitor its surface quality in the manufacturing process automatically. Considering that it is difficult to achieve the automatic separation of complicated surface defects with mutual interference, a novel method is proposed to adaptively determine the center point of clustering and separates defects through using the density peaks and spectral multiple-manifold clustering(SMMC). First, for the extracted binary image of the surface defects, the principal component analyzer of mixed probabilities is used to estimate the local tangent space of each pixel. In addition, the similarity matrix between the local tangent space of each defect location is improved. Then, to overcome the problem that SMMC must manually determine the number of clusters, an improved method based on clustering by density peaks is presented to determine the clustering center point according to the similarity matrix. Finally, the pixels included in each principal component analyzer are assigned to different defect manifolds through SMMC. The surface defect locations ultimately achieve separation and detection. The experimental results show different defects with mutual interference on the surfaces can be automatically separated and more accurately detected.