rega anggraini's picture

Dear All,
Is it possible to Acid Value predictable by NIRS ?
I found something different from our calibration for Acid Value, the result of Est. Min is zero.
Can you give me some advice why our results is like this, because our calibration or any other reasons?
Is it "Est. Min" has influence for our result? or something like that can be ignored?
Thank you so much for your help.

Uploaded Images: 
hlmark's picture

Rega - I don't know if you can create a calibration for Acid Value. In fact I had to look it up to find out what it was, which turned out to be "# mg of KOH needed to neutralize the acid in a sample" (my paraphrase). I see several possible problem areas:
1) The sample contains many different acids, each of which could give a different spectral response in the NIR. Do you know what types of acids you might expect to be in the samples, and what their spectra are (this applies only to organic acids)?
2) The sample contains only inorganic acids (e,g,, HCl, H2SO4), none of which would be expected to have an NIR response. We would also expect the KOH to not have any NIR spectrum.
3) The sample contains both types of acids, in which case the amount of acid is highly unrelated to the spectrum.
4) Conceivably we might expect the water spectrum itself to be affected by large changes in H+ or OH- concentrations, which might provide a basis for an NIR measurement, but in this case the spectral changes would be severely non-linearly related to the Acid Value (e.g., pH would have a logarithmic relationship), whereas all of the common calibration algorithms require linear relationships between the spectrum and the analyte. The non-linearity could be in the X (spectral variable) or the Y (analyte value) values; the two cases need to be handled differently. Draper and Smith's "Applied Regression Analysis", Wiley (any edition), includes good discussions on handling each type of non-linearity. Although written for MLR, their advice about handling non-lienarity can apply to all calibration algorithms. For a more recent NIR example (with non-linearity in the analyte value) and how it was handled, see Appl. Spect.; 64(9), p.995-1006 (2010)

ianm's picture

Dear Rega,

do you want to determine the acid valid of a polymer (e.g. polyester, polyamide) ?

If yes, let me know and I can send you an example.

Kind regards



Prof. Dr. H. W. Siesler
Department of Physical Chemistry
Schuetzenbahn 70
D 45117 Essen
E-Mail:  [email protected]

(posted on behalf of Heinz Siesler by Ian Michael)

ptillmann's picture

This is obviously an WinISI specific question. It regards the on-screen information of WinISI. Rega is stumbling over the "Est min" entry, which is 0.0 and therefore a quite unusual value for avic value.
don't worry. The 0.0 has no implication on your prediction.
It was calculated by WinISI as follows: max(mean - 3* SD , 0.0). You can check this with the "Est max", which is calculated as "mean + 3* SD".

rega anggraini's picture

Thanks for your answer Peter,​
I want to confirm your text : It was calculated by WinISI as follows: max(mean - 3* SD , 0.0). "Est max", which is calculated as "mean + 3* SD".
Does "max" incorrect, be supposed is "min" ?
So, Est. Min is (Mean - 3*SD) and Est. Max is (Mean + 3*SD).
I understand for your explain, and i have next question about it.
Does it mean that the Est. Min value is not the minimum value inputted in
And, do you mean that Est. Min is the minimum value shown by NIRS prediction?
Could you please explain it?
Best Regards,

ptillmann's picture

The entry on-screen "Est min" is calculated by the WinISI software as follows:
- calculate Mean -3* SD (these values can be taken from the above table in your uploaded picture)
- if the result is < 0.0, than Est min: = 0.0
Correct: the "Est min" is not the minimum value in the calibration set, hence "estimated". :-)
"Est min" has no effect on you prediction.

rega anggraini's picture

Thank you Peter.
Nice to discuss with you.
I hope you can help me again if i have other problem about WinISI :-D