Application of Convolutional Neural Network for Classification and Tracking of Weld Seam Shapes for TAL Brabo Manipulator

Abstract With ongoing advancements in robotics, robotic welding has gained popularity in industries since it reduces human intervention in hazardous and unpleasant working environment. Welding is a process in which two workpieces are joined together. The edge interface of the two halves are called weld seam. The main scope of this paper is online detection and classification of different shapes of weld seam from webcam using 2D Convolutional Neural Networks (CNN) in MATLAB. The problem with real time detection of weld seams accurately is that images contain high noise levels that are difficult to process leads to reduced tracking accuracy with traditional image processing techniques. To surpass this problem, this paper presents a low cost solution to detect and trace weld seam using CNN which has the powerful feature extraction capability and images are captured using webcam. This proposed methodology can be used to detect any weld seam shapes and classify automatically. To detect the weld seam, 1000 datasets of different weld shapes have been created with noise using 2D vision sensor and it is pre trained by 2D convolutional technique. Using this pretrained CNN model, live weld shape images are captured and classified according to the different shapes making the robot to track corresponding weld seam path. Experimental analysis is performed in 5 DOF TAL Brabo manipulator in which servomotors of the robotic arm are controlled by Trio motion controller which can communicate with MATLAB through ActiveX COM object. Experimental validation has been done based on probability scores and accuracy.