A Robotic Passive Vision System for Texture Analysis in Weld Beads

摘要

The use of robots is increasing in different industries, as in the case of large metal structures. The use of mobile robots meets the needs of these industries: it can be easily moved in the production line, with gains in repeatability and process efficiency, reducing rework costs. The poor configuration of the robotic and welding system generates problems in the internal structure and the surface of the weld, thus compromising the final quality of the piece. Visual inspections are common to identify problems that may have occurred during these processes. Inspections that use x-rays, ultrasound, or thermal cameras require additional equipment and highly trained inspectors to analyze the results and detect problems. This work presents a computer vision system based on a passive monocular camera for analysis of weld bead textures. Images of weld beads with and without discontinuities are captured and a dimensional reduction algorithm known as Principal Component Analysis (PCA) is applied to select the main characteristics that describe each group. Afterward, the Support Vector Machine (SVM) supervised learning method is applied to recognize the patterns of image groups, making it possible to classify new images of weld beads as welds with or without discontinuities. The system makes use of the same camera coupled to the robot responsible for conducting the welding, without the need for additional sensors, and assists the welding inspector in the evaluation of the performed process.