АВТОМАТИЧЕСКОЕ РАСПОЗНАВАНИЕ ПОВЕРХНОСТНЫХ ДЕФЕКТОВ СВАРКИ ПРИ ПРОИЗВОДСТВЕ ГИБКИХ ТРУБ ИЗ НЕРЖАВЕЮЩЕЙ СТАЛИ

Abstract

The paper describes the automatic recognition of surface welding defects on video stream frames while manufacturing flexible stainless steel pipes. The defects that appear during automatic tungsten inert gas welding are studied. A universal algorithm based on the background subtraction method is presented for detecting and localizing various welding defects on video stream frames. A solution to the problem of recognizing such defects as lack of fusion and scaled surface is also proposed. A dataset containing texture features of defects was created. Classifiers were trained using the support vector machine. Experimental studies have shown that the proposed solutions allow to prompt detect any deviations occurring during metal welding and recognize certain welding defects on the processed frames of the video stream.