Charpy Curvefit 1

[brubakerj]

Hi,
I was hoping to get some feedback for the following applet I put on my website

I am performing non-linear regression on Charpy data and calculating the parameters.

http://www.novanumeric.com/samples.php?CalcName=Charpy

Any advice/direction/feature requests/constructive criticism would be greatly appreciated!

-Jeff

 

[salmon2]

Your example fits almost perfectly. Are they the real testing data? Let me guess, is it a normalized steel, if they are real?

Seems like you need four data points to do the fitting. That is a lot. Ideally I hope to just use two or three at most. Otherwise, it does not make sense to do the numerical fitting any more.

I guess the heat treatment or microstructure affect your function selection. Are you fixed with tanh function? I guess you will have to use some other functions. For example, the FCC austenite does not have the sharp drop at all (or until very low temperature).

I also suggest you to do further regression to link some of your constants with physical parameters. For example, grain size has a general positive effect on the toughness. I bet one of your parameters will have a linear relationship with grain size.

Possibly you can find one profile for each different microstructure. Further normalizing couple parameters if you can will make this idea a pretty good research paper.

 

[arunmrao]

You have sufficient data points to form a curve. I hope the curve is not drawn first and then points placed to substantiate the inferences.Recently I reviewed a technical paper ,and a curve fitting was done with just 2 data points!!

_____________________________________
"It's better to die standing than live your whole life on the knees" by Peter Mayle in his book A Good Year

 

[brubakerj]

The data I grabbed from a paper describing the curve fitting procedure, but I don't think it's real data. I'm coming at this from a numerical method perspective, and not from performing experimentation. I was hoping to get some real world examples to try. Is one or two data points practical? Everytime I've worked with metallurgist, they do seem to draw trends from a few data points, is that normal for a metallurgist?

Would you mind sharing those data points with me?

Regards,
Jeff Brubaker

 

[stanislasdz]

salmon2 :

Great answer !

thank you for your contribution

 

[TVP]

JB,

Usually impact testing involves more than 2-3 data points; probably 4-5 is the norm. It is easy to find actual data on this: here is a Google search that shows tables and graphs of impact results that include discrete measurements:

http://www.google.com/search?q=char...576l53779l0l55419l7l6l0l0l0l0l187l718l2.4l6l0