Measurement of weld penetration for variable-groove weldment using dual-band imaging and neural network

Abstract

For the process of gas metal arc welding, the online measurement of weld penetration has always been a research hotspot, and high-precision penetration monitoring can lay the foundation for penetration control. This study developed a novel method for measuring the weld penetration of variable-groove weldments using dual-band imaging and a neural network. First, a welding temperature field measurement system was constructed using a colour charge-coupled device (CCD), and the feasibility of using the output images from the red and green channels of the CCD for colorimetric temperature measurement was investigated. Then, a welding temperature field measuring algorithm based on a network model was investigated. Compared with the traditional temperature measurement algorithm, the proposed algorithm can measure the welding temperature field with higher precision. The distribution characteristics of the temperature field were extracted by means of the histograms of oriented gradients (HOG) algorithm. Finally, based on a back-propagation (BP) neural network, a weld penetration prediction model for variable-groove weldments was built, and the distribution features of the temperature field were used as the model input. The experimental results reveal that the proposed back bead penetration measurement method for variable-groove weldments can achieve the high-accuracy penetration measurement, which has broad application potential.