![](/mwf/attachments//avatars/3040-3756.png)
You're "Weldcome" nantong.
![grin :grin:](/f2/epx.png)
I concur... This discussion has stimulated my curiosity further... So of course, I wasn't satisfied with the direction I was following in attempting to answer your question:
"Henry is there a neural solution for this?" In a previous post in the thread, I had already asked if anyone has been accepting, or even adopting the Bayesian Neural Network for modeling and prediction of ferrite number in a previous post... And you replied by asking me the question above... After some research and further reading, out of the void that sometimes constitutes part of my memory on any given day... Somehow I remembered about ORFN (Oak Ridge Ferrite Number) that also used neural network analysis... And is an improvement to the FNN-1999 model...The FNN-1999 model, developed with the same data as the WRC1992 constitution diagram, was shown to reduce prediction errors by as much as 40%... I will quote a couple of paragraphs from the article I found in my library and then looked up online for anyone to read...
"Furthermore, composition dependent effects of alloying elements were taken into account and the consequences of this were clearly demonstrated... A more recent model, which is also based on neural networks, also demonstrates significant improvements compared to the WRC1992 constitution diagram... Another major shortcoming that is present in the traditional constitutional diagrams as well as the newer models, including the recently developed neural network models, is the absence of any consideration of welding conditions and how they may influence the ferrite content of welds... In particular, the weld cooling rate will have a significant effect on the final ferrite content...There are two ways in which the cooling rate will influence the ferrite content: 1) The cooling rate will alter the extent of the diffusion-controlled transformation of ferrite to austenite during cooling in the solid state, and it may influence the solidification behavior." Well, going back a few years ago, I did remember reading an article in the back of the Welding Journal that was titled:
"Improved Ferrite Number Prediction Model that Accounts for Cooling Rate Part1: Model Development Details of a prediction model based on a neural network system of analysis are described BY J.M.VITEK, S.A.DAVID, AND C.R.HINMAN"http://www.aws.org/wj/supplement/WJ_2003_01_s10.pdfThis is from the original article: "SUMMARY and CONCLUSIONS: A new model (ORFN©) that takes welding conditions into account when predicting FN of stainless steel welds has been developed. Several simplifications and assumptions were required during the development of the model. However, the new ORFN© model represents the first prediction model that quantitatively accounts for the effect of weld conditions on FN. It has been shown that the ORFN© model correctly predicts the variation in FN due to solidification model changes and suppression o the solid-state ferrite to austenite transformation at high cooling rates.The ORFN©model is particularly useful for high-speed welds, duplex stainless steel welds,and high-power density process welds."
http://web.ornl.gov/~webworks/cppr/y2001/pres/113776.pdfhttp://web.ornl.gov/~webworks/cppr/y2001/pres/113800.pdfHere's a more recent paper in the AWS Welding Journal:
http://www.nxtbook.com/nxtbooks/aws/wj_201204/index.php?startid=118-s#/146I think I'm going to make a comparison analysis of the two models (Bayesian & ORFN) in order to find out which one is more accurate... I'll let you know what results & conclusions I come up with.
![roll :roll:](/f2/epx.png)
Oh wait! It's already been done in the Bayesian paper!
![surprised :surprised:](/f2/epx.png)
![lol :lol:](/f2/epx.png)
Now I'm done.
![cool :cool:](/f2/epx.png)
Respectfully,
Henry