Improving the tracking quality of the weld seam butt with V-form grooving by using Kalman filter and neural network

Abstract

An algorithm for tracking of the welded seams grooving by using a Kalman filter based on six characteristic points of the profile obtained using the RF627 laser vision sensor is proposed. In order to reduce the error in weld seams control, a multilayer neural network with a backpropagation algorithm is created to compensate for errors caused by colored noise when using the Kalman filter. Experimental results show that when the algorithm is applied, the error in tracking the trajectory of weld seams is reduced. Keywords tracking of weld seams; multilayer/multi-pass welding; Kalman filter; multilayer perceptron