There are many ways of measuring precision in GD-OES. A common method is to measure intensities repeatedly at the same place. This is known as 'internal precision'. An example is shown below for a stainless steel sample (465/1) measured four times. The experimental conditions were: RF power 70 W, pre-integration time 240 s, and integration time 15 s. The experiments were performed using a JY 5000 RF, Horiba Jobin Yvon, Longjumeau.
The first column shows the element and wavelength (nm), the second column the mean intensity ratioed to Fe 386, the third column the RSD %, and finally the four individual measurements. Fe 386 and Fi (ie, total light) are shown with * because they are the direct intensities, ie not ratioed.
Another common measure of precision is: 'external precision' where the same sample is measured repeatedly in different places. For homogeneous, well-polished samples the external precision approaches that of the 'internal precision'. If the external precision is significantly greater than the internal precision on the same sample, then the sample surface may need repolishing or the sample may not be homogeneous.
Short/Long Term Stability
Instruments may change slowly over weeks or months or years. These small, long term changes can be corrected, for example, by cleaning the window in the source, or by recalibration (also called drift correction).
But an instrument should be stable in the short term, e.g. over the time it takes to make a series of measurements, such as a calibration, i.e., over a period of at least some hours.
The GD source is a highly stable source and optical spectrometers are stable devices so the combination should give good short term stability.
Short term stability is a good test of the performance of an instrument. Instruments are routinely tested for their short term stability during production. An example is shown above where the intensities from a stainless steel sample have been measured repeatedly during a day to check any changes in intensity with time:
Clearly this instrument was performing relatively well. Wether it is good enough or not depends on the requirements.
First published on the web: 8 October 2000.
Authors: Richard Payling, Thomas Nelis