Neural network modeling of weld pool shape in pulsed-laser aluminum welds

Abstract

The submitled mpnurcript has been aclhored by a contractor d the U. S. G o w r~ under coruract NO. OE-AC05-660%?2464. Acootdingly. the U.S Government retains a nonexdwve., royaly-tree license lo puMish or reproduce !he pubiished form of this contribrtbn, of alow others to do so. for US.

Modelling correlation between alloy composition and ferrite number in duplex stainless steel welds using artificial neural networks

“…Thus, depending on the base alloy composition, the role of the alloying elements varies. Similar observations were reported by Vitek et al 12 The role of Ni, N, Mn, and Cu as austenite stabilisers is in agreement with reports by Kotecki. 3 The role of Si and Nb as an austenite stabiliser in alloy 255 is reported here for the first time.…”

“…The size and shape of the error bar in each figure depends on the distribution and noise of the training dataset used. It was found that C, Ni, and N act as austenite stabilisers, with N having a stronger effect than C. This is in agreement with the results reported by Vitek et al 12 Although the Mn content in the training dataset varied from 0?5 to 1?98 wt-% for duplex stainless steel welds, the variation in Mn content was found not to influence the FN. The observation that variation in Mn does not influence the FN is in agreement with that reported by Szumachowski and Kotecki 22 for austenitic stainless steel welds.…”

“…The FNN-1999 model, in which Nb content was assumed to be zero (as it was not available), reports that variation in Nb concentration does not influence the FN value. 12 Figure 2 also shows that Cr, Mo, and V act as ferrite stabilisers, while Ti and Co have varying effects. Titanium acts as an austenite stabiliser at low Ti contents ,0?05%, but acts as a strong ferrite stabiliser at Ti contents .0?05% due to the formation of nitrides that remove the potential austenitising effect of N. At low concentration levels Ti is not effective in forming carbides and nitrides and so C and N remain in austenite and that in turn stabilises the austenite.…”

“…The correlation coefficient obtained was 0?98, which is better than for any of the currently available FN prediction methods. 12 The generalised BNN model for FN prediction 20 also revealed the significance of individual elements on the FN value. It was found that the FN value was influenced insignificantly by Mn and Nb, most significantly by Cr, Ni, Mo, and N, and less significantly by C than by N. More importantly, it was found that the elements Si, Ti, V and Co also significantly influence the FN value, elements which are not included in the WRC-1992 diagram.…”

“…The method relies on linear relations, which may not be justified. 10 Neural network analysis involves nonlinear regression analysis 11,12 and hence is better suited to complex cases involving many possibly interacting variables. 13 In neural network analysis, the data are presented to the network in the form of input and output parameters, and the non-linear relationships are found by minimising the difference between measured and predicted values.…”