Content-based X-ray image analysis of aluminium castings

Abstract

The X-ray digital images of aluminium castings containing different microstructure defects were characterised by feature vectors composed from: 1 first-order statistics 2 singular values 3 second-order statistics calculated from grey level co-occurrence matrices (GLCM). The most suitable features were found by means of the Ward’s clustering method. The X-ray images characterised by the first-order statistics, such as 1st to 6th statistical moments and entropy, were portioned in two main clusters with efficiency of 90%. Consequently, using the six statistical moments and entropy, aluminium castings were sorted according to their quality in comparison with one casting of no observable defects. Their similarity was expressed measured by the Euclidean distance (ED). At ED = 5, the quality aluminium castings were effectively separated from the defective ones. This image analysis approach can be simply implemented into the automatic quality control of metallurgical processes and could be also used for the retrieval of similar microstructure defects in image databases.