Dear All
As we know, to determine the detection limit for AAS /ICP, we can do 10 blank tests and then calculate the standard deviation, so that LOD and LOQ can be calculated. My question is how to determine the detection limit for Bench top XRF machine? the matrix is iron ore
Your cooperation will be highly appreciated. Thanks
Regards
Azlan
Ask your question to a search engine such as Google. There are many answers to this question.
E.g.
https://www.researchgate.net/publication/228687395_Detection_limit_and_estimate_of_uncertainty_of_analytical_XRF_results
The variance of the blank is only one method of calculating LOD, and in my opinion not a very good one. I vastly prefer calculating LOD from the analysis of low level standards. For example, in your iron ore, you certainly have standards that you use for all the elements you are looking for. Multiple measurements of the lowest level standard will allow you to calculate the statistical variance at the level in your standard (it's even better if you have multiple measurements over time). From this data you can calculate average reading and standard deviation, which will allow you to calculate LOD using various statistical methods (I use the United States EPA method found at 40 CFR 136 Appendix B). Obviously from the LOD you get the LOQ since it is defined in terms of the LOD.
XRF is a counting technique. You can use counting statistics to compute the best possible detection limit. A counting of N photons results of a lottery draw from a Poisson distribution centred on N. This can be very well estimated with a normal distribution centred on N with a variance of N. You measure a peak with a bulk count Np (variance Np), you also measure the background Nb (variance Nb), so the net peak is Nn = Np - Nb. The variance of Nn is the sum of both Np and Nb variances (additivity of variance for independent normal distributions), equal to Np+Nb. The standard deviation of the net peak is then the square root of the variance, sqrt(Np+Nb). If you are close to detection limit, Np~Nb, so the standard deviation on the net peak is ~sqrt(2Nb). Your detection limit is therefore 3 times STD, so is 3*sqrt(2Nb). Often instruments give intensity I instead of counting N, you have to adapt using: N = t*I where t is the counting time.
That is the best LOD possible. If your instrument is unstable, this add a excess variance that deplete your LOD. To test this issue, you can draw observed variance of a given signal (expressed in N) that should be proportional to the counting time t. Or the variance of a given signal in I, that should be proportional to the inverse of t.
There is a staristical metod based on the square root of the srandard deviation. In your determination use 6 concentration levels and three replicates at each level.
- Badges
- Science topic