Process monitoring and adaptive quality control for robotic gas metal arc welding

Abstract

The aim of this research was to develop an adaptive quality control strategy for robotic
\ngas metal arc welding of thin steel sheets. Statistical methods were used to monitor and
\ncontrol the quality of welds produced.
\nThe quality of welds cannot be directly measured during welding. It can however be
\nestimated by correlating weld quality parameters to relevant process variables. It was
\nfound sufficient to do this using welding current and voltage transient signals only.
\nThe strategy developed was problem solving oriented with emphasis on quality
\nassurance, defect detection and prevention. It was based on simple algorithms developed
\nusing multiple regression models, fuzzy regression models and subjective rules derived
\nfrom experimental trials.
\nThe resulting algorithms were used to
\ncontrol weld bead geometry;
\nprevent inadequate penetration;
\ndetect and control metal transfer;
\nassess welding arc stability;
\noptimise welding procedure;
\nprevent undercut;
\ndetect joint geometry variations.
\nModelling was an integral part of this work, and as a feasibility study, some of the
\nmodels developed for process control were remodelled using "Backpropagation"
\nArtificial Neural Networks. The neural network models were found to offer no
\nsignificant improvement over regression models when used for estimating weld quality
\nfrom welding parameters and predicting optimum welding parameter.
\nAs a result of the work a multilevel quality control strategy involving preweld parameter
\noptimisation, on line control and post weld analysis was developed and demonstrated
\nin a production environment. The main emphasis of the work carried out was on
\ndeveloping control models and means of monitoring the process on-line; the
\nimplementation of robotic control was outside the scope of this work. The control
\nstrategy proposed was however validated by using post weld analysis and simulation in
\nsoftware.