https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html https://www.euro-online.org/enog/inoc2007/Papers/m https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html

How to improve accuracy in curve fitting

shenl024's picture
Forums: 

In atomic emission spectroscopy, the concentration of the unknow sample is determined by the calibration curve from the known sample.
curren accracy is about 5%. how to improve it to 1%?
 
Thanks.

hlmark's picture

Hello, shen024. Several immediate questions arise

1) What kind of fitting curve are you using, and how are you calculating it?
2) How are you determining "accuracy"? This question includes both the measurements you make and the calculations you apply.
3) What is the nature of the sample?
4) If you are comparing the instrumental value to a nominal "true" value, how do you determine the "true" value?
5) What spectroscopic technique are you applying: transmission, reflection, ATR, etc.?
"A chain is only as strong as its weakest link" applys here. You have to look at every part of the determination of "accuracy" to make sure it's working correctly.
Howard
\o/
/_\
 

td's picture

Hello Sheno024,
You have come to a NIR forum; but we always attempt to answer all questions!
It is 30 years since I was last involved with Atomic Absorption (AA) spectroscopy so methods may be very different!
Our claibration graph was made using a series of anlyses using known concentrations of the analyte of interest. and the software would fit a straight line through these points. In 1996 an MSC student of mine (Michael Coene) was looking at methods of comparing NIR spectra by regression when there were large variation in the particle size of the materials investigated. He discovered that quadratic regression often gave improved results compared to linear regression.(1) This might be useful for you to try for AA?
This does not diminish the warnings that Howard gave you. One of the important problems with AA is the effect of the sample matrix on the mesured absorption and this can give rise to errors considerably greater than 5%. For this reason I preferred to use the method of standard additions. [Adding standard amounts of the analyte to the sample and obtaining the result graphically] I suppose this might be what you are doing?
Ref 1   Cone, Grinter & Davies, JNIRS, 4,153-161 (1996).
Best wishes,
Tony Davies