Optimization of welding parameters in underwater wet FCAW on a structural steel using support vector regression and sequential quadratic programming

The underwater welding process used to repair offshore engineering structures involves a large number of processing parameters that must be selected and strictly controlled to achieve the required metallurgical and mechanical characteristics. Internal and surface porosity is a very common serious problem in underwater welding directly dependent on the process parameters. In this research work, a novel modeling and optimization study applying support vector regression (SVR) model and sequential quadratic programming (SQP) algorithm in order to predict and optimize the main process parameters of underwater flux cored arc welding (FCAW) on surface porosity in bead-on-plate welds of ASTM A36 steel is investigated. For this purpose, the experimental underwater wet FCAW process was carried out in an open tank with a 30-cm fresh water column. A composed central design was used to obtain the experimental matrix of 17 underwater welding tests with several combinations of parameters and variation levels: voltage ( V , 22–26 V), welding speed ( v , 3–10 mm/s), and wire feed speed ( WFS , 6350–11,430 mm/min) and the surface porosity was counted to each weld bead. In general, the results showed that this new computational approach using SVR satisfactorily approximates the level of surface porosity according to the operational variables. The model was validated with experimental results obtained from optimization with sequential quadratic programming algorithm. Using both mathematical methods, it was possible to obtain adequate parameters ( V = 24V, WFS = 8890 mm/min, and v = 3.4 mm/s) to experimentally produce underwater welds with minimal surface porosity (fewer than 10 pores in a bead length of 30 cm).