In addition to the ASD, RSD,LOD and SR, stability is a performance feature of AE spectrometer. Spectrometer reading are subject to changes with time, being affected by following phenomena；

  CCD instability

  Temperature changes

  -mechanical and electrical properties

  Shock

  -optics

  Contamination

  -optics

  Vacuum

  -absorption of radiation

  Supply gases (composition, state)

  -Ar for sparking, contamination, flow rates

   Geometric changes

  -electrode trip, deposit at injector top.

  When carrying out calibration with calibration samples, nominal values for each calibration curve are on recalibration samples from a number of measurements. Although the function of the calibration curve does not change, its position in the co-ordinate system changes due to changes in sensitivity and in the zero point of the spectrometer. Later measurements from recalibration samples give actual values which are compared with the nominal values:

  Recalibration sample T: nominal_T=a+b actual_T

  Recalibration sample H:nominal_H=a+b actual_H

  The error in the mean value declines with n measurements. Numerous measurements should therefore be taken for recalibration eg. 6, or as many spark burn spots as can be placed on the surface of the recalibration sample.

  From the 2 equations given above, “a” and “b” are determined and applied to the actual intensities or intensity ratios of the analysis samples as follows:

  Nominal=a+b·actural.

  It should be mentioned that, in the case of metal analyses with SD, the low sample is often the pure metal. The high samples are generally synthetic compositions with as many analytes as possible having concentrations in the upper half of the calibrated concentration ranges. The compositions have to comply with metallurgical considerations, so that it is not possible to have any analytes at will in any concentrations in one and the same sample-and, finally, an appreciable quantity of base element should be still be present.

  Particularly in analytical lines which tend towards self-absorption, recalibration must not be carried out with concentrations above the calibrated range but must performed at half the concentration. For example, with SDAR , for Mg 285.2nm c_max 0.3%; Sr 460.7nm c_max 0.3%; Be 313.0 nm c_max 0.1%.

  How often should recalibration be performed? To answer this question it is necessary to know the stability or instability of the radiation-measuring machine, and have a definition. It is assumed that spectrometers with SDAR are often used in an 8-h shift cycle, and about 100 samples have to be analysed, so that about 300 sparking operations are carried out.

  Measurements with recalibration samples on a large number of type 31000 spectrometer gave the following information:

If 300 measurements are carried out in any desired sequence within 8 h, and in each case 10 individual values measured in direct succession are summarized as a mean value m, the following applies for intensity rations:

  m_max - m_min<3%

  It was already known at that time that all absolute intensities dropped with increasing number of sparking operations but to a variable extent, so that the intensity ratios I_EI/I_Ref dropped, rose or remained the same, depending on whether the measurement values for the analyte channel dropped less, equally or more than those for the reference channel.

  The above formulation of stability has been retained to this date. Then as now there were particularly stable or less stable intensity ratios which could not be assigned either to an analyte or an analysis line or spectrometer number.

  Now, as was also the case then, CCD are changed if the stability requirement is not met. If better stability values are required, particularly for analytes which have to be accurately determined, a number of CCD must often be tested. Stability guarantees better than 2% are still unrealistic since the problem lies not with the spectrometer manufacturer but with the CCD manufacturer, whose name for all of them is… Japan. Suspicion about CCD being responsible for instability has been reinforced by the fact that when using CCD in an spectrometer with constant radiation, stability values about 3 times better than with SD are measured.

  The stability of CCD was tested under radiation conditions such as are present with spectrometers using SDAR:

  The instability values are between 94 and 99% of the initial values. Depending on the stability values of the CCD for the analytical line and reference line, instability thus may be 0-5%. The only remedy for this nuisance so far is selection of CCD or recalibration.

   Stability tests carried out in accordance with the above schedule. Nothing of a systematic nature can be detected with respect to the analyte or reference line.

  To answer the question of how frequently recalibration should be carried out, the following procedure is proposed:

  After stability measurements, with replacement of CCD if necessary, have led to the required results, calibration is carried out in-plant; the spectrometer is delivered and commissioned. Recalibration is carried out once every 8 h.

  If samples with analytes are to be measured where the possible instability is acceptable, a check sample must be measured which has a known chemical composition and which should be similar to the analysis samples. Actual value deviations from the nominal values are added up. The sample to be analysed is assumed to eg. a CrNi steel with about 20% Cr and 10% Ni. The check sample gives:

  Nominal 21.00% Cr  Actual 21.20% Cr =+0.2%Cr

       9.85% Ni         10.00% Ni=+0.15% Ni

  For the analysis sample the following is found:

  Actual 19.20% Cr after correction Nominal 19.00% Cr

      10.45%Ni                     10.30%Ni

  This cumulative calculation can only be carried out when the analysis sample and the check sample have similar concentrations for the analytes. Otherwise the actual values must be added to the nominal values by multiplication and the factor used on the analysis samples until the check sample is measured again or recalibration is carried out. This use of check sample is also known as type recalibration. Today corrections are calculated automatically.

  Tolerances ranges must be set for the actual values of check samples. A programme can used to detect variations in actual recalibration values or actual check sample values against time, so that it is possible, shortly after commissioning, to state “rules of behaviour” for recalibration and checking a spectrometer.

  The check sample method does not improve long-term instability but only “over-comes” it. This method leads to good analyses but is time-consuming and inconvenient when a wide range and sample qualities has to be analysed daily.

  There are different views about methods of drift correction, one of which will be presented here:

  A distinction must be made between “basic recalibration” and “check recalibration”. In the case of basic recalibration the recalibration samples are measured 6 times; in the case of check recalibration this is often limited to duplicate measurements.

  Since with basic recalibration the recalibration coefficients “a” and “b” are more reliable than those obtained with check recalibration and experience indicates that they alter only slightly within short time intervals, the mean value from the coefficient so far used and from the coefficient calculated during check recalibration, a stop watch should be started and should be read-off for each check recalibration. For the b value of a measurement channel the following record may be obtained.

  According to this record, for analyses which lie between the 2^nd and 3^rd.check recalibrations, the measurement values are to be converted with b=0.995. The drift during the day should be calculated at the end of a working day. Using the data from the record given as an example, this has been done for the first 4 h. The “b” value increases by about 0.27% per h ie . The sensitivity of the channel declines about 0.27% per h. If this is linked up the next day with drift from the previous day, a permanent drift or stability check is automatically obtained.

  When measuring a fairly large series of samples, the following procedure should be adopted:

For the recalibration samples the recalibration coefficients “a” and “b” are calculated immediately and stored, together with the time. For the analysis samples and the check or recalibration samples incorporated in these, only the measurement values are stored, together with the time. If the series is small, the measurement cycle can be repeated. At the end of the measurements the drift functions of the individual channels are first calculated, and after this the measurement values for the analysis samples are converted using the appropriate coefficient of the drift functions. In the example in Fig. the measured valued for an analysis sample, measured at the time 2.25h, is converted with b=0.986+0.00267*2.25=0.9929.

  What has been stated here for the coefficient b , also applies in the same way to a.

  As a result of recalibration, a large quantity of numerical data is obtained within a few weeks which is suitable for reliable determination of the RSD. I f these data are followed-up, changes in the RSD will then give an early indication of incipient changes in parts of the spectrometer.

  Recalibration calculation are carried out automatically by the spectrometer computer. The recalibration makes it possible to calculate what intensities or intensity ratios the spectrometer would have given for any sample for any sample if this sample has been measured at the same point in time at which the nominal values for the recalibration samples were determined. Recalibration thus has nothing to do with calibration. A spectrometer can be recalibration any number of times and thus held ready as a “radiation-measuring machine” without having to carry out a calibration.

  One-and two-point recalibrations are the procedures. It is wrong to try to carry out a recalibration on more than two points in order to compensate in this way for changes which occur in the curvature of calibration lines in the course of time. In this case, the CCD is operating outside and not inside its linearity range… which should be the aim.