Experimental qualification and physics-informed predictive modelling of robotic and manual MAG welding of S355MC steel

All robotic and manual welding procedures were tested on S355MC steel using standard specimens and were produced as per the EN ISO 15614-1 requirements for three welding procedures, i.e., Robot MAG HD plus (with pulsing frequencies ranging between 100 and 400 Hz), Robot MAG (short circuit), and Manual MAG. There were no defects present in any of the robot welds during the testing phase (with a rule-of-three 95 % upper constraint of ≤ 8.3 %) as against the 27.8 % defect rate detected in manual welds (i.e., 5 of 18 specimens). For Charpy impact energy, the standard deviation in robot welds was 4.6-5.1 J, compared to 19.5 J in manual welds. The process capability indices for robot welds were CpL = 2.57 (Robot HD+) and CpL = 2.54 (Robot SC), whereas for manual welds, CpL = 0.71; the 95 % bootstrap confidence intervals for the two weld types did not overlap. The optimised GRNN surrogate model results related to cross-mode validation are found to have high values of R2 (0.9987-0.9994) and low values of RMSE (1.8-3.2 MPa). The R2 and RMSE were established with the same (h = 0.04) smoothing parameter (kernel bandwidth). The Arc-Entropy Stability Index (AESI), which is a physics-based measure, corresponds with the Charpy coefficient of variation (Pearson r ≈ 0.99, n = 3). In order to confirm these findings, a thorough validation was conducted employing a range of methods, including XGBoost Regression, Monte Carlo Reliability simulation, Retrospective Power Analysis, and detailed UQ. Optical metallography on analogously welded specimens corroborates the inferred acicular-ferrite/ferritic–pearlitic weld metal and the absence of displacive constituents. The relationships reported here are predictive and correlational.