Ductile Iron nodule count with software VS charts

Hi,
We've report nodules count by comparison with ASTM A247 charts since many years. Typical values with ASTM A536 Ductile Iron was from 100 to
200 /mm2.
Now we are looking to use an image analysis software that generates many data. One of those is nodules count. We had a big surprise that for our same samples, the software is reporting from 400 to 500 /mm2 and we are interpreting with the charts only 100 to 200 /mm2.
My first try was to eliminate more small particles with the soft. The results was now less then 400 /mm2 but very far from what we are expecting. What I think I understand is that those ASTM charts use samples that have almost one size of nodules. In real life, there are no samples that can show this type of results. We always have a mixed population of nodules sizes and shapes. When we try to choose the right nodules count with the chart we must filter a lot what we see. Software doesn't do that. It computes what it sees and it's better like that.
Am I the only one who has problems with this?
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
snipped-for-privacy@yahoo.com (phil) wrote in message

Phil,
Same thing is happening with grain size ASTM E112(and inclusions). ASTM has not yet published bias and reproducability, repeatablity, or reliablity of automatic grain size determination using round robin testing. Personally I don't trust a software engineer to write an algorithm to cover even the basic requirments that human with a theory of operation instinctively understand.
First, you must verify that everything is set up right: Is the magnification that the software USES is the same you use? - are all calculation using the manual (your) and automatic (software) correct?
Second, validate a series of different samples manually and using the software you have then and find someone with other software or contract a lab to make a third party measurement using automatic and manual methods. Statistically compare using ANOVA or Gage R&R methods (check AIAG or ASQ for avaiable text on statistics applied to measurments).
Ed Vojcak PE (do not reply to my E-mail it is a dead end)
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Edward D. Vojcak wrote:

A long time ago, early 1980's, the company I was working with bought a Lemont Image Analyser. This was a hardware thing that would work with image scans.
It claimed to calculate amazing things.
I wanted it to calculate amazing things to help move quantitative microscopy up to where I could use it in theories of mechanical behavior.
We had a lot of technicians who had never, or only a few times, measured anything from the tens of thousands of micrographs that they had taken (SEM and optica microscopy).
The results the machine was giving from real samples were hard to believe.
Therefore, I made up some test images using my drafting equipment and my skills with India Ink.
In other words, we knew almost exactly what the answers were supposed to be.
Sadly, the Lemont machine badly failed to give these answers.
I published them in a Technical Report.
Lemont threatened to sue.
A few years later, Lemont was no longer in the business.
One of the guys admitted that in the rush to get the hardware out the door and sold, the checking of the software was of real low priority.
So, Ed Vojcak's advice is excellent, even 20 years after this fiasco.
Don't trust a software engineer ........
Jim
--
...............................


Keepsake gift for young girls.
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
The ASTM/DIS charts are fudged to a large extent. (email me for a detailed explanation).
If you actually perform an analysis of the sample images you can get significant variation in results. This is because the results are operator and software dependent. Image prep, methodology used for evaluating acceptance criterion, etc. all affect the results. The majority of software packages out there are "research tools" geared towards a PhD metallurgist. These packages tend to be too powerful in that there are so many options it makes consistency between operators/shifts etc. impossible. You are dependent on good training of your techs to make it work for you.
Try the following link: www.nodman.com
True 1 button analysis, operator independent, 100% repeatable. Works on the shop floor. Made by a foundry guy to work in a foundry environment.
Cheers,

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

BTW...What is the gage R&R of this system? What do you mean by 100% repeatable? is this repeatablity within the sample or between samples from the same cast?
Ed
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Phil, Feature extraction software will do it, the image data will need to be preprocessed using spatial operators, in addition to convolution filters, you most likely will need to use some morphology functions like binary or grayscale erosion and dilation, this will better control the dimensional geometry issues. After this you may use feature extraction to determine whether the carbon inclusions are truly nodular. To do this you may look for small major and minor axis ratios for validation. I should also suggest using the special "CDL" illuminator, its available from "NER" as well as other vision lighting suppliers, good lighting greatly simplifies the everything else. Alan Black

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here. All logos and trade names are the property of their respective owners.