A inodcl was dcvelopcd to prcdict thc wcld pool shapc in pulscd Nd:YAG lascr welds of aluiiiinuin alloy 5754. Tlic inodcl utilizcd ncural nctwork analysis to rclatc tlic wcld proccss conditions to four pool shape par,lmctcrs: pciictration. width. width at half-pcnctriition. and cross-scctional i i m. Tiic iiiodcl dcvelopmcnt involvcd tlic idcntification of tlie input (proccss) variables, tlie desired output (sliape) variablcs. and tlic optiinnl neuraldictyo~k $ccl@gyc:: ?The lattcr was influcnccd by tlic numb~~f-acfined=Idputs;andJoulputs as ivcll as tlic ainount of &?la that ?ysgvailable for training tlie nctwork. Aftcr appropriate training, ilid%cst":det?&& was identificd and was uscd to prcdict tlie weld shape., &routine to convert tlic shapc parainctcrs into prcdictcd ~tcld~-pr~fiEg;was also dcvclopcd. This routiiic was bascd on tlic actual csperimcntal weld profilcs and did not iinposc an artificial analytical function to dcscribe tlic wcld profilc. Tlic ncural nctwork iiiodcl was testcd on cspcriinciitiil wclds. Tlie iiiodcl prcdictions wcre csccllcnt. It was found that tlic prcdictcd shapes \\ere witliin tlie espcrimcntal wriatioiis t1i:it wcrc found along the lcngtli of tlic welds (due to the pulscd nature of the wcld powcr) and tlie rcproducibility of welds IiIiidC under noiiiiniilly identical conditions.
A new semi-empirical model for predicting the ferrite content of stainless steel welds has been developed. This model predicts the ferrite number of stainless steel welds as a function of composition. The model is based on an equation representing the free energy change between ferrite and austenite. This model has been derived from published data of experimental weld metal compositions and their corresponding ferrite numbers. The predictive capability of this model was found to be good and describes the effect of alloying elements on the ferrite number. This model is comparable in accuracy to currently available constitution diagrams but is applicable to a wider range of alloy compositions.
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.
Prcdicting the fcrrite content in stainlcss stccl welds is desirable in order to assess an alloy's susceptibility to liot cracking and to estimate the as-welded propertics. Scveriil methods liave been uscd over the years to estimate the ferrite content as a function of the alloy composition. A new technique is dcscribcd which uses a neural network analysis to deterinine the ferrite numbcr. The network was trained on tile same data set that was used to gcnente tlie WRC-1992 constitution diagram. Tlie accuracy of the ncural network predictions is coir pared to that for the WRC-1992 diagram as well as another recently proposed method. It was found that the neural network model was approximately 20% more accurate than either D f the other two methods. In addition, it is suggestcd that furthcr improvements to the ncural network model. including the consideration of process variablcs. can be made which can lcatl to even better accuracy.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.