Neural network based approach for quality improvement of orbital arc welding joints

Neural network based models are developed and used for the description of the relations of the geometry characteristics of Steel 3 welds from orbital arc welding (OAW) process parameters. This integrated methodology is implemented together with response surface methodology (statistical approach) for the investigation of the defined as quality characteristics: outer and inner weld widths. Both implemented modeling approaches are compared and their applicability is discussed. Regression models are estimated and neural networks were trained using a set of experimental data containing different welding regime conditions (pipe diameter and thickness, welding current and time for one full turn of the electrode). The implementation of both approaches and their applicability for process optimization and automatic control aiming improving of the quality of the obtained welds is compared.