An integrated approach for feature extraction and defect detection in industrial radiographic images - case study on welding defects

Abstract

Non-destructive testing (NDT), through radiographic image inspection is widely used in industry for ensuring the quality of the manufacturing processes. Generally this test explores defects in welding, casting and moulding processes. Conventionally the inspection processes are carried out manually; recently automated systems are developed to improvise quality of the manufacturing processes. In such system, automatic image interpretation and classifications plays a major role. When such tasks are hand-engineered, it calls in for human experts making the system more constrained, and there arises a need for developing an automated system to address the issue of the NDT process. Several researchers proposed statistical and geometrical-based approaches for defect classification in weldment. However, the results from these methods had significant misclassification rate. In order to reduce the rate of misclassification during the detection process, a novel approach is proposed in this research paper. The approach incorporates fractal-based image analysis for obtaining features from an input image and a fuzzy-based rule engine for detection and classification of circular/longitudinal defects that appears in the radiographic image.