International Journal of Engineering and Technology

Pattern recognition of weld defects can be done by partitioning the data points into clusters. The weld defect arises due to various types of adverse conditions and carelessness at the time of welding which leads to major disasters at the later stage. This paper proposes a method for detection of Burn Through, Slag Inclusion and Lack of Penetration weld defects in radiographic images which follow a specific pattern for each type. The image is preprocessed using Histogram equalization and segmented using Region growing methodology. The feature dimension reduction is performed using PCA, which is an unsupervised dimension reduction methodology which chooses the dimensions with the largest variance. K-Means clustering is used for clustering a cohesive group from a set of patterns after performing dimensionality reduction using Principal Component Analysis. The result of classification using this technique is providing a very accurate classification of weld defect.