Spectrometer - check samples (C)
With increasing demand for quality assurance (ISO 9000) it is also necessary to have reliable analyses by spectrometer. An efficient spectrometer will repeatedly find the same analysis value within the limits preset for drift and precision (eg. 2s of the precision value). There is a commonly held view that the efficiency of spectrometers must be tested with certified reference material (CRM).
A spectrometer is calibrated using a large number of reference samples, some of which may be CRM's. Testing of spectrometer calibration using the same reference samples or even the same CRM's is commonplace since it is of course based on these samples.
It is advisable to include in the spectrometer calibration process a large number of reference samples (production quality samples) which fully represent the qualities of the analysis samples, because AE-spectrometry is still arelative measurement process with matrix - and spectral interferences. CRM's are mostly produced by synthesis and do not correspond to the matrix of the analysis (production) samples either with respect to composition or sample structure (history). (See E DIN 51009, page 17, 1995). For this reason CRM's may deviate systemetically from the calibration curve valid for the analysis samples, so that calibration testing for AE-spectrometers using CRM's is only possible unless it is established with certainty that their material properties are in conformity with those of the analysis samples. It is very difficult, of not actually impossible, to ensure that this is so, unless the same calibration is found.
To ascertain the efficiency of AE-spectrometers, spectrometer-check samples should be used (see E DIN 51008 - 1, page 38, 1995). The only requirement is for precision comparable with the recalibration samples; their exact analysis is not asked for. In contrast to recalibration samples, spectrometer-check samples have compositions corresponding to the analysis (production) samples or have such low alloy levels that they lie on the calibration curves produced with reference samples. In contrast to CRM's (<20 mm high) the dimensions of spectrometer-check samples are those which are optimum for routine use with AE-spectrometers (40-60mm high, 40-50mm diameter). Two to three times higher than CRM's, they are 2-3 times cheaper so that their cost/performance-ratio is 5-10 times better. Expenditure (work + time = cost) for CRM''s makes them quite unsuitable for testing the efficiency of AE-spectrometers. In addition, with CRM's the aim is to accommodate numerous elements over the widest possible concentration ranges within a set comprising as few samples as possible (eg. 5 pieces). As a result, there is often a large difference in composition compared with analysis (production) samples, particularly for high-alloy qualities, so that the calibration curves are not always identical.
Spectrometer efficiency can be checked by means of intensity values or concentration. For the sake of clarity, it is usually done with concentrations, as follows:
After the spectrometer has been calibrated with reference samples the spectrometer-check samples are analysed a number of times (min. 6 times), immediately or subsequently (in a reliably recalibrated state) ie. "integrated" with the calibration obtained from reference samples, so that they themselves become reference samples.
Using spectrometer check samples, the efficiency of a spectrometer can be testet for eg.:
a) SPC quality assurance
b) when the spectrometer analysis of an analysis (production quality) sample is not to be trusted
c) at regular intervals or at specific stages in a production cycle
(eg. finished melt sample, every 100% piece of a product)
d) to check the need for recalibration. The spectrometer-check sample is then also known as the spectrometer drift check sample.
Note: For "Rolling calibration" see: Recalibration samples.
Method d) is particularly to be recommended when measurements are taken by spectrometer on only a few samples per day and these samples belong to only a few quality groups (for example they are only low-alloyCrNi steel, only GG and GGG cast iron, only AIMgSi, only CuZn, only Zamak, etc.)
The elements in the spectrometer-check sample/s must be allocated sensible tolerance limits from the values determined by spectrometer.
The actual values for calibration determined with recalibration samples and the actual check values determined with spectrometer-check samples must be taken and assessed by statistical methods, leading to greater operational reliability of the spectrometer, while providing information on the actual stability and frequency of recalibration, thus also saving material for recalibration samples.
Notes on "Integration of spectrometer-check samples into calibration"
the calibration curve is established using a number of reference samples (eg. 30-50) whose (true or probable) (chemical) analysis value has been properly established according to the rules of good analysis practice.
Examining the analysis values obtained from a large number of laboratories (at least 10) selected for a Round Robin test for this purpose (and this is necessary for CRM's), it will be found that these laboratories do not give one and the same value. With a sufficiently large number of analysis values, they give a normal (Gauss) distribution with a standard deviation, which, in the most favourable cases, is about 1% of the mean value ie. eg. for a cast iron sample with 3,50% C, it presents a single standard deviation (1s) of ± 0.035% C. At 1s, 68% of the values are within range. Assuming, as is commonly stated for acuracy indication, 2s values are so that 96% of the values are with range, this means (3,50 ± 0,07)% C or (3,43 - 3,57)% C as the result of the Round Robin test.
All values found within this range for a single test (laboratory, test room etc.) are statistically possible and thus comparatively accurate.
If theerefore the reference samples do not all lie correctly on the calibration curve (and, for this reason, numerous reference samples are taken for calibration so that it is possible to determine an equalisation curve) it cannot be expected that a single spectrometer-check sample will lie on the curve with it is (probable) analysis value.
Let us assume that a cast iron sample has a (probable) nominal value of 3,50% C. The analysis from the spectrometer for which the sample is to be used as the spectrometer-check sample gives 3,45% C.
What can be done???
It is not permissible, under any circumstances, to alter the position of the calibration curve (eg. by altering the high point nominal value) because this curve has been established with a large number of reference samples which, for their part, may lie up to ± 0,07% C away from the equalisation curve (calibration curve).
The value of 3,45% C (sinceit agrees with the nominal value of 3,50% within ± 0,07% C (2s)) can be used as the nominal value for checking the servicability of this spectrometer. Permitted tolerances eg. (3,45 ± 0,03)% C (for min 6 measurements) must be assigned (according to the precision and stability of the spectrometer) to this (spectrometer) nominal value. If the check value (actual value) is within (3,42 - 3,48)% C, the servicability of the spectrometer is assured.
NB: ± 2s of precision is often taken as spectrometer tolerance (in the above example ± 0,03% C).
