Automatic Quality Monitoring in GMA Welding using Signal Processing Methods
Sammanfattning: An ongoing continuous process of automatization of the production lines is used in industry to reduce production costs. Automatization of quality control should be seen as part of a cost reduction process, as well as of quality control of the welding. Mostly the time-consuming and expensive checking procedure of the weld joint is carried out off-line by an experienced welding operator. The need for automatic quality control is thus great in order to achieve higher cost efficiency.The present thesis focus on the weld quality of pulsed, short-circuiting and spray Gas Metal Arc welding. The purpose is to design algorithms to detect defects on-line and automatically during a robotic welding pass using the measured weld voltage and current. Experiments with different specimens are performed in order to cause defects in weld joints and introduce symptoms in the measured signals. The defects produced in the weld joint are found to be of two main kinds. In pulsed welding, burn-through occurs. In spray arc and short-circuiting welding, the size of the weld was reduced. The two types of defects and their relation to the measured welding voltage and current are discussed. The environment around the measurements is highly disturbed by electrical noise. Several measurement techniques is used accordingly in order to prevent interference of the measurements. To obtain a reliable detection, features which are sensitive to changes in the weld process must be extracted from the measured signals. Features which are suggested in other works of this field, and features which are obtained through observation of the measured signals from the experiments of the present thesis are investigated. The investigation shows that the mean and variance of the weld voltage and current yield promising results when welding in short-circuiting and spray mode due to the robustness of the extracted features. When welding in pulsed mode, the prediction error from an autoregressive model with extra input is used as a feature. The extracted features are fed into a detection algorithm to further enhance the signal to noise ratio of the resulting test quantity. The filtering of the features is then compared to a threshold. Six statistical detection algorithms for weld quality have been designed and tested on measured data. Two methods have been applied to pulsed GMA welding, one for short-circuiting welding and three for the spray arc welding. The results obtained from the experiments shows that it is possible to detect changes in the weld quality both automatically
Denna avhandling är EVENTUELLT nedladdningsbar som PDF. Kolla denna länk för att se om den går att ladda ner.