X-ray fluorescence is generated by, among other processes, the ionization of the inner shell of an atom. If ionization of an inner shell does occur, i.e., an electron is removed, then the hole is filled by an electron from a higher energy shell. The energy difference is released as X-ray fluorescence radiation and is characteristic for the element. XRF analyzers use the X-ray fluorescence radiation released to determine the composition of a sample.

In practice from fluorine to uranium, 72 elements can be analyzed with XRF; almost the entire periodic table of elements, which is the reason for the widespread use of this technique.

With EDXRF, the sample is excited by the X-ray tube directly or through a filter. A semiconductor detector analyzes the X-ray fluorescence radiation that comes directly from the sample.

Here, the detector together with the associated electronics counts and sorts, according to energy, all of the photons that reach it. A pulse height spectrum that indicates the number of photons or impulses for a given energy is established. The detector typically has only a few µs for processing, so that processing is accordingly limited to approximately 1,000,000 pulses per second. Using a filter, a portion of the exciting radiation can be screened out to avoid overloading the detector.

The major difference to EDXRF is the method of detection of the X-ray fluorescence radiation for WDXRF. Using a goniometer, only one wavelength from the spectrum is fed to the detector, i.e., it measures only one line from one element. In order to conduct multiple element analyses, it is necessary to create a serial measuring program that drives to and analyzes all of the lines of interest; one after another. There are, however, so-called simultaneous spectrometers: In this case, there is a set channel for each element consisting of a fixed crystal with corresponding detector arranged around the sample. When combined with a goniometer this forms a very fast, high performance XRF instrument that is especially useful for process control.

In general, it is possible to say that the two techniques are complimentary; one supplements the other. EDXRF has a time advantage, as all elements are measured simultaneously, whereas the (serial) WDXRF measures the elements one after another. WDXRF has a resolution and sensitivity advantage which is especially useful in the range of atomic numbers up to 30 and from 55 to 80.

With XRF, most of the elements are only measured on the surface of the sample. This is why the condition of the sample surface with respect to smoothness and homogeneity (surface area and depth) is so essential. In practice, this more or less determines the total analytical error. Thus sample preparation for XRF becomes the most important component of a test method.

There are five typical sample preparation forms:

• Solid sample prepared as a loose powder
• Solid sample prepared as pressed powder with or without binder
• Solid sample in its original form
• Liquid sample in a sample cup
• Solid sample prepared as a fused bead

CRM: Certified Reference Material
A CRM is a RM that has been analyzed with a traceable analytical method for one or more parameters. The value of the parameter, the corresponding uncertainty and a statement concerning the metrological traceability are indicated in a certificate.

RM: Reference Material
A sample that is sufficiently homogenous and stable for one or more parameters and that is considered to be suitable for use in a measuring process.

SeRM: Secondary Standard
A sample for which the parameters were assigned using a comparison with a primary measurement standard (e.g., a CRM).

XRF is characterized by a very high long-term stability. Nevertheless, the components show certain wear effects over time so that the measured signal changes. If the change is known, it can be corrected for using a so-called drift correction.

The drift correction is based on the principle that at the time of the calibration (day 0) and at a given time afterwards (day 1), the same, stable drift correction sample (drift monitor) is measured. The correction is then calculated from the change in the measured signal.