Validation of X-Ray Fluorescence
EDXRF Benchmarking Using Analytical Performances
X-Ray Fluorescence is a powerful alternative to Inductively Coupled Plasma Optical Emission Spectrometry technique. A benchmarking of EDXRF instruments was launched to maintain awareness about their analytical performance capabilities and for allowing up-to-date recommendations to factory end-users. An EDXRF benchmarking methodology is explained with the description of the different steps, the critical points to respect in order to ensure a proper validation and the key performance criteria on which to focus.
Nowadays, X-ray fluorescence spectrometry is a well-established analytical technique for quantitative elemental analysis of various samples . Good throughput, simple usage, cost savings and possibility to directly analyze powdered samples are important features that have made it a mature tool for routine quality controls in various industries. XRF technology is currently implemented in multiple food production sites, for process control and release of finished products (e.g. infant cereals, milk-based powders). Benchmarking has been identified as an essential way to maintain awareness about the latest EDXRF instruments and their analytical performances, allowing up-to-date recommendations to XRF end-users located in factories.
Several years ago, the Analytical Methods Committee  listed the features of interest such as Intrumental criteria (excitation, detection and sample handling) and non-instrumental ones (supplier services and data treatment). This list stays theoretical – indeed, the data collection is tedious and is also strongly dependent on the good will of suppliers and quality of provided information.
The methodology for EDXRF benchmarking is described in steps, the points to respect in order to ensure a proper validation and the key performance criteria on which to focus. The important quantity of work and data generated was directly obtained for one of the applications of interest (macro and trace elements in milk-based powders) and was used for later internal recommendations.
Methodology and Trials
The benchmarking study was divided into several steps:
Expected specifications and interactions with suppliers
The specifications/expectations were the following:
- Elements and concentration ranges
- Experimental conditions (medium, analysis time, sample preparation, auto sampler and spinner)
Suppliers were informed about the activities (analytical method development and support to factory end-users), the benchmarking methodology and expected specifications.
The defined rules were the following:
- Same sets of samples submitted to each supplier,
- Transparency on the performance assessment results, without naming the other participants.
Most of the suppliers have accepted to participate to this benchmarking.
Milk-based powder was selected as a matrix of interest. For EDXRF, samples were prepared in our laboratory by pressing 3 pellets of 6 grams.
For reference method concentrations, acidic digestion of samples was performed using high pressure microwave prior to analysis of Na, Mg, P, K, Ca, Mn, Fe, Cu and Zn by ICP-AES. Chloride was assessed by potentiometry. Both methods are official AOAC methods. All samples were analyzed in duplicate.
Method Development and Samples Analyses
The 22 calibration samples, for which the reference concentrations were communicated to suppliers, were analyzed in triplicate and used to calibrate the instruments. The 28 validation samples which were supplied without reference concentrations, were analyzed in triplicate and considered as unknown samples. The method development and analyses were run by the application laboratory of each supplier, defining the appropriate instrumental operating conditions and calculation models.
Data Treatment and Evaluation of Analytical Performance
The first step was the assessment of quantification limits which were obtained by analyzing:
- Recovery of obtained EDXRF values versus reference ones (macro elements range = 90-110%, trace element range = 80-120%)
- Relative Standard Deviation between replicates of same sample (macro elements max = 5%, trace element max = 10-20%)
This process allows to evaluate the instrument capability to accurately measure the concentrations in the lower part of calibration range for each element. Only when limits of quantification values were found to be appropriate, the analytical performance was assessed.
The evaluations were performed using average concentrations of the replicates analyzed by reference methods and by EDXRF for each sample. Evaluation of the EDXRF performances was done using robust statistics. These performance characteristics were internally calculated to ensure a strict data comparison and application of the same statistical formulae previously described in a previous paper .
For each analyte obtained on the 8 instruments, by identifying the minimal value of SEC, SEP, SEP/SEC, SD(d), SD(r), SD(iR) and the maximal value of R2, it was possible to calculate normalized performance characteristics for each instrument. Then, a performance index was established for each analyte and each instrument, as the average of the normalized performance characteristics.
Results per Element
The case for potassium is given below on Table 1. The longer a segment is for a performance characteristic, the better it is. For the SD(r): Instrument B is found to be the more repeatable out of the 8 instruments with a value of 2 mg/kg, whereas instrument D is the less repeatable with a value of 11 mg/kg.
Results per Instrument
The Figure 3 shows that instrument B is able to properly cover the ten investigated analytes that the instrument G cannot analyze Na, Mg and Mn (for which, limits of quantification were found too high). Model G is able to analyze only seven other analytes, with generally lower analytical performances than instrument B (except for Cl).
This benchmarking presents numerous advantages, including increased efficiency and the saving of time and money. On the supplier’s side, the tests performed allowed them to learn about new applications and to have access to samples with well-defined contents in minerals. On your side, this benchmark was conducted without renting or even purchasing instruments, as method development and validation analyses were run in supplier laboratories. Nevertheless, this approach could be challenged as incomplete for the following reasons: the non-instrumental criteria linked to data treatment, software, supplier services/reliability, spares availability, quality of documentation and training are all of major importance but cannot be really assessed prior to purchase, installation and regular usage on site.
In total, 8 instruments were compared using their analytical performance characteristics. This comparison led to the following up-to-date recommendations:
- identifying of best candidate(s),
- obtaining full validations instead of one-day demonstration on site with too few samples,
- avoiding redundant work to be done by each interested factory or by each contacted supplier.
Minerals and Imaging group, Nestlé Research Centre
 Taylor A., Barlow N., Martin P., Day M., Hill S., Patriarcae M. and White M.: Journal of Analytical Atomic Spectrometry 31, 554-596 (2016); DOI:10.1039/C6JA90005D
 Analytical Methods Committee: Accreditation and Quality Assurance 11(12), 610–624 (2006) DOI:10.1007/s00769-006-0187-8
 Perring L. and Blanc J.: Food Analytical Methods 1(3), 205–213 (2008); DOI:10.1007/s12161-008-9030-7