Research On Deviation Recognition Based On Rough Sets And Rotating Arc Sensor For Narrow Gap MAG Welding

As a highly effective welding method, rotating arc narrow gap MAG welding has beenwidely used in thick plates joint. In order to ensure the welding quality and avoid weldingdefects such as wall incomplete fusion, the arc need to be in the groove center during thewelding process. So the efficiently real-time seam tracking control must be carried out,where the key is to recognize the welding deviation reliably in the real time.Rotating arc sensor was used for narrow gap MAG welding in this paper. To study therotating arc’s electrical signal changes, workpiece was designed and processed to mimicmulti-layer single pass welding groove at first, and enough experimental data were acquiredunder different deviations in the paper. Second, the data were analyzed, and it showed thatthere were relations between the extreme values’ distribution and the welding deviation.Then, this relation’s rationality and reliability were verified by Matlab simulation andprogramming. Therefore, considering the characteristics of welding and algorithm’sreliability, a welding seam tracking algorithm was put forward based on the number anddistribution of welding current’s extreme. The algorithm only concerned with the extremevalues’ distribution and was not sensitive to the adverse effects such as groove’s processingerrors, assembly errors, etc. In addition the algorithm was also simple and better real-time.Furthermore, in order to realize intelligent control of narrow gap MAG welding seamtracking, the seam deviation was modeled based on rough sets theory. First, according to thesimulation results of arc length and variation of experimental data, the current signals weredivided into different intervals separately in each rotating cycle. The mean value of eachinterval and differences between left and right intervals were computed and introduced ascondition attributes to build decision tables. Then after discretization and reduction fordecision table, the knowledge model of “If…Then” form was obtained. The model’sprediction ability was validated and compared with BP neural network model. It showedthat both models had similar predictive capability, and RS model precise could meet actualneeds. Further more, RS model had better comprehensibility, and was useful to findpotential laws between seam deviations and welding electrical signals from experimentaldata. The research was helpful for controller design.