Passive machine vision-based defect classification in tungsten inert gas welding on SS304 using AI-based gradient descent algorithm

Abstract

In modern digitization, safety industries demand flaw-free and high-integrity welds, due to part localization on high uncertainty makes automation a challenging task. Integrating robotic welding with high-value manufacturing sector makes volume rise through pre-programmed repetitive performance on desired welding space. Further, the establishment of fully automated robotic-based welding operations with different configurations on controlled localization is the least available due to unviable cost. Hence, the proposed work concentrates on internet of things (IoT)-driven robotic tungsten inert gas (TIG) welding on stainless steel (SS) 304 by incorporating online programming (OP) with visual control schemes to classify the nature of weld quality using an artificial neural network. Continuous sensory-guided techniques with IoT high-level operator interface affords automated welding planning by using feature matching strategies. Area scan-based complementary metal oxide semiconductor (CMOS) cameras have been used to capture passive vision real-time images of the weld species for defect classification. In real-time, discrete reference workpiece and image feature extraction processes seem to be complicated in an unstructured welding environment. Hence, the present idea will enhance the efficiency of high product variance, but the accuracy of the automation relies on the weld image database. The proposed method (a) adaptively regulates its welding parameter variation on the welding trajectory path, (b) adapts and produces precise good weld workpieces with a 92% production rate with flaw-free welds and (c) initiates the automatic tuning of robot kinematics through IoT closed-loop online external control strategy. The proposed research experimental results confirm that fully automated IoT-driven robotic TIG welding affords good welding with an 88% improved quality rate through online passive vision-based image feature analysis.