Beginners Guide to ISE Measurement, Chapter 14.

RECENT RESEARCH INTO CORRECTIONS FOR INTERFERING IONS.

NOTE: This was a development project carried out in 1998/9 which was never brought to full commercial viability due to lack of interest from potential users. Also it was realized that it would never be cost-effective because of the huge amount of effort required in constructing the calibration data base which would probably need repeating for each ISE/reference electrode combination.

Also see: A new Approach to Ammonium Analysis Using Ion-Selective Electrodes

The correction for interfering ions has always been a problem in ISE measurement. The traditional way of dealing with this problem was to use chemical methods to try to remove the interfering ion from solution by complexing, precipitation, or the use of specific ion-exchange resins, but most of these methods are either tedious and time consuming, or expensive. The other alternative was to measure the concentration of the interfering ion first and then make a correction based on the published (or experimentally determined) selectivity coefficient for the electrode concerned. As noted previously, this could be difficult and time consuming, and relatively unreliable and imprecise because the selectivity coefficient is variable and dependent on a number of different factors including the concentration of both ions. In many cases, the presence of interfering ions means that the ISE method cannot be used for these analyses.

The development of the ISE computer interfaces now gives the possibility of simultaneously monitoring both the primary ion and the interfering ion and using sophisticated data processing techniques to make the appropriate corrections. However, initial attempts to develope this technique revealed a number of difficult problems. The first system to be considered was the measurement of ammonium in the presence of potassium. After evaluating various mathematical techniques, it was decided that artificial neural network software would be the most appropriate, but it was soon realised that this would require a large calibration data base in order to work effectively. Several groups of standards would be required to define a four dimensional net with a range of ammonium concentrations each mixed with a range of potassium concentrations, plotted against their respective millivolt values. This immediately presents difficulties because of the electrode potential drift problem.

As discussed earlier, if a group of solutions are measured one after the other and then the first one is measured again it will normally give a significantly different result from the original value. In normal ISE potentiometry this is overcome by using only a small number of calibration standards so there is little drift between the first and the last measurement, and recalibrating frequently between small groups of samples. However, a different approach is required in order to build up a large calibration data set.

A further potential problem with interference corrections which had to be addressed at the outset was the variability of electrode responses. Initially it was anticipated that each ISE/reference electrode combination would have to be calibrated separately and this could be very time -consuming and costly. Fortunately, subsequent testing of various production batches of electrodes demonstrated that, at least for the ELIT ammonium and potassium ISEs used with a lithium acetate reference, the electrode responses are sufficiently uniform that a single calibration can be used for all combinations. Thus it should be cost effective to spend considerable time and effort in building up a large data base to make the correction for potassium on ammonium.

It was found that two techniques could be used to overcome the problem of drift during calibration. The easiest method is to repeatedly re-measure the first standard before every other standard measurement and then normalise the new value to be compatible with the first value measured. For example, if a subsequent reading of the first standard gives, say, 5 mV lower than when first measured, then 5 mV must be added to the results for the following new standard, and so on for each new standard solution. Alternatively, a group of samples can be measured one after the other at regular time intervals, and then re-measured in the reverse order. The average of the two values will then be compensated for drift over the period of measurement. The advantage of this method is that each standard is measured twice thus increasing the reliability of the data. If a second group of calibration data are needed to extend the range of the first then these can be measured on a separate occasion, in the same way. In this case a single common standard must be used in order to normalise the second batch of data to the first. This latter method was adopted in order to create a calibration net for the ammonium-potassium interference.

Initially, twelve standard solutions were made containing 10, 20, and 30 ppm K each mixed with 1, 2, 5, and 10 ppm NH4 and measured twice as described above. The ammonium and potassium electrodes, together with a lithium acetate double junction reference, were plugged into a triple electrode head connected to a two-channel interface, and the millivolt readings were recorded simultaneously. Each millivolt measurement was the average of ten readings taken at one second intervals in a still solution, two minutes after immersion of the electrodes and swirling of the solution. The electrodes were rinsed with a jet of de-ionised water and then soaked in water for 20 seconds between each standard solution in order to minimise hysteresis effects.

These data were then plotted on a graph of mV NH4 versus mV K and lines were drawn through points representing equal concentrations of either element i.e. the mV graph was contoured for concentration. This produced a calibration net from which the concentration of both elements could be interpolated after plotting the millivolt values for an unknown sample. These data demonstrated that the basic principle for making interference corrections was valid, but also demonstrated the need for much more data in order to define the shape of the net more accurately, and to extend the range of concentrations. It also showed that only those samples that plotted within the boundaries of the net could be measured and that this simple graphical approach would be very difficult to interpolate accurately. Thus there was a need for a more sophisticated, computer-based method for interpolating the data.

In order to improve and extend the calibration, more standards were made so as to define the external boundaries of the net. This time, 1 and 100 ppm NH4 were each mixed with 1, 5, 10, 25, and 50 ppm K and 1 and 50 ppm K were mixed with 2, 5, 10, and 25 ppm NH4. In addition, one solution containing 10 ppm NH4 and 20 ppm K, from the previous calibration batch, was re-measured so that the new data set could be normalised to the old, and hence plotted on the same graph.

Once the boundaries of the net and a few internal points were defined, it was then possible to define more calibration points from the cross-over of the contour lines, and so extend the calibration data base without having to spend even more time in making-up and measuring new solutions. These data were then entered into special software to form the data base for training the neural network so that it could be used for interpolating the concentration of measured samples.

It must be noted that this data base was determined for pure standard solutions without the addition of ISAB. Hence the slight curvature of the boundary lines at the highest concentrations, as the effect of the increasing ionic strength becomes more apparent. Thus this calibration is really only appropriate for measuring relatively pure sample solutions where the ionic strength is likely to be similar to that of the standards of similar concentration. If complex samples with high ionic strength are to be measured then it will be necessary to construct a new data base using standards with added ISAB which will not show this additional curvature at high concentration. In this case, all sample solutions will also have to be mixed with the same proportion of ISAB before measurement.

Specially written software has been developed to incorporate the neural network module and the ammonium - potassium calibration data base so that these can be used to measure the concentrations in unknown samples containing both ions. The first task to be performed each time the system is used is to recalibrate the data base to allow for day-to-day variations in the electrode responses due to differences in temperature or ageing of the electrodes. This is done by re-measuring one of the standards used to make the initial calibration, preferably one nearest to the expected range of concentrations of the samples to be measured. The original data line (containing ppm NH4, ppm K, mV NH4, mV K) is selected from a table of the standard values and the original solution, or preferably a freshly made solution with the same composition, is measured. The new data for the millivolt response for both electrodes is compared with the original. If these are significantly different then the whole calibration net can be normalised to these new values by adding or subtracting the differences in the millivolt readings. This modified calibration data is then used for re-training the neural net before measuring samples. The original and new values for the re-calibration standard are recorded in the sample results table.

Samples are measured by simply immersing the three electrodes (ammonium and potassium ISEs and lithium acetate reference electrode housed in a triple connector electrode head) and waiting for a stable reading. The on-screen display shows the millivolts for each electrode and the concentration of both ions in ppm and mole/l. When the readings are stable the results are saved in the data table together with the date and time of measurement. Sample numbers or comments can be added by the operator. In order to compensate for electrode drift during measurement, a further re-calibration can be made quickly and easily at any time, by simply returning to the re-calibration screen and re-measuring the single re-calibration standard. These new data will also be recorded, chronologically, on the results table. The data table can be printed directly or exported to other software packages.

PREVIOUS PAGE | BACK TO CONTENTS LIST