Automatic detection of welding defects using radiography with a neural approach

Abstract

Radiography is a method of evaluation and non-destructive control. Given the relevance of radiographic inspection of many different areas of industries, there has been numerous research projects aimed at automating the analysis and the interpretation of welding discontinuities. In this work, the automatic control and inspection of welding defects is made by edge detection method of radiographic images, based essentially on the use of a Multilayer Perceptron (MPC). This paper describes an origin method to detect welding defects existing in the weld radiography based in the use of the artificial neural networks, aim the classification and the increase of the successful recognition default percentage. Therefore this work is done in four steps: the first step consist in the preparation of the database to be used later in the second step for training MPC. The next step is done in two main parts; the first part is devoted to the detection of contours, then the second section consists in the elimination of additional contours. Finally, it concludes with on real cases study.