Calculaion of Measurement Uncertainty in Pipette Calibration with QMSYS GUM Software Stefan Golemanov QUALISYST Ltd., Bulgaria Email: Stefan@qsyst.com Abstract Pipette calibration is the most essential part of every laboratory, and as such, requires careful understanding and selection of calibration methods and procedures in order to prevent errors, deviations and inconsistencies. This article provides an example for uncertainty analysis in pipette calibration with all relevant steps in equation modeling and uncertainty calculations. Furthermore, the paper examines two approaches for the relevant calculations: by means of a professional software and an excel add-in program. The presented data is supported with graphical representations and illustrations. Keywords uncertainty analysis, pipette calibration, gravimetric method, measurement uncertainty calculation, GUM, GUM Uncertainty Framework, Monte-Carlo method analytical balance the amount of pure water that is purged from the piston pipette to a specific vessel and to convert the obtained measuring result in mass unit into volumetric unit. In practice, a number of 5 to 10 repeat measurements are made to determine the systematic and random errors of the calibrated pipette [1]. To the semeasurements, corrections must be applied in order to compensate for any deviation from standard temperature, atmospheric conditions and any significant evaporation of the water during the calibration. Variable volume pipettes should be tested at three points over their designated range: maximum (nominal) volume, 50% of nominal volume and the lower limit of the volume range or 10% of the nominal volume, whichever is the greater. 1. Introduction Volume measurement is an important step in most industrial and analytical measurement operations. Piston pipettes are used for very precise measurements in many fields such as chemistry, health, biology and pharmacy. In order to reduce and identify possible errors in liquid handling, it is necessary to calibrate pipettes using the correct methods. National Metrology Institutes and accredited laboratories are using as standard method the gravimetric method to calibrate volume instruments. In this article, this method will assist the calibration procedures as well. The basic principal of the gravimetric method is to weigh with Lab World Magazine 16 Vol. 03 No. 03 Feb-April 2014 It is also necessary to evaluate the measurement uncertainty, since this must be stated with the result of the measurement to give the end use confidence in the measurement. Some uncertainty contributions vary with each tested volume, pipette channel, calibration equipment and environmental conditions. Therefore, each pipette calibration must be accompanied by individual calculations of the measurement uncertainty for every testing volume and pipette channel. In practice, the measurement uncertainty calculation is a complex process, because when calibrating a variable onechannel pipette, the measurement uncertainty should be calculated 3 times separately for every measured volume. For multi-channel pipettes, the number of uncertainty calculations is multiplied by the number of the channels – 24 times for 8channel pipette, 36 times for 12-channel pipette. The complexity of the measurement uncertainty analysis often requires the additional use of a software product. The example analyzed further in this article is aided with the QMSys GUM software, which demonstrates practical and easy, yet accurate and fully reliable method for calculation of the measurement uncertainty in pipette calibration. Figure 1. Developing the measurement model and specifying the input quantities In the basis equation of the gravimetric method [1], [2] are included two additional quantities, describing the influence of the temperature differences between water, air and pipette (δVT_DIFF), and the operator contributions (δVOP). 2. Measurement Uncertainty of the Pipette Calibration by the Gravimetric Method The correction for the thermal expansion of the pipette is given by The systematic procedure of building measurement uncertainty analysis according to the GUM Uncertainty Framework consists of the following stages: ● Formulation: this stage consists of defining the output quantity, identifying the input quantities, developing a measurement model that relates the output quantity to the input quantities and assigning probability distributions to the input quantities on the basis of available knowledge ● Calculation: the measurement uncertainty calculation is performed by using the appropriate methods: -1 where αc is the cubic expansion coefficient in °C , tp is the pipette temperature in degrees Celsius;tp20 is equal to 20 °C. The cubic expansion coefficient is not generally valid determinant due to the different design and construction of each pipette types. Material properties, material combinations and different geometries and types affect the cubic expansion coefficient. However, these influences are not mathematically representable and therefore cannot be defined by all manufacturers. It should also be emphasized that the temperature of the piston-pipette is not constant because of hand-warming at the middle and the top, and because of the evaporation-cooling process at the bottom of the pipette. The cubic expansion coefficient is approximated with the value "zero" and the correction for the thermal expansion is not included in the model equations. -GUF method for linear models (GUM Uncertainty Framework) -GUF method for nonlinear models with symmetric distribution of output quantity -Monte-Carlo method for all models and distributions of the result quantities Summarizing and Validation: this stage summarizes and validates the results and prepares the uncertainty budget. When calculating the uncertainty with the software, the user needs to furnish only the formulation stage, while the program performs all other stages of the uncertainty analysis. ● The equation of the weighing result consist of input quantities, describing the repeated measurements of the delivered water, the correction for evaporation loss, the calibration of the balance and the rounding error due to balance resolution. The uncertainty contributions due to the finite resolution of the balance are listed twice in the model –once for the zero-setting and once for the measurement at load. 2.1. Developing the Measurement Model The formulation stage of uncertainty evaluation involves developing a measurement model, incorporating corrections and other effects as necessary. The model equations (Figure 1) are the starting point for all subsequent calculations by the software and are developed in accordance with the DKD guide for pipette calibration [2], which offers the most detailed analysis of the factors, influencing the measurement result. Vol. 03 No. 03 Feb-April 2014 Air and water densities (ρA,ρW) are calculated by the approximation formulas by Jones and Haris, given in [4]. The equations for the environmental conditions (air temperature, air pressure, water temperature and relative humidity) follow 17 Lab World Magazine the same model and consist of input quantities, describing the indication of the measuring instrument, the calibration of the instrument, the rounding error due to finite resolution and possible drift during the pipette calibration. The following table presents all quantities of the measurement model: * Quantity type: R – result, IR – interim result, A – type A uncertainty estimation, B – type B uncertainty estimation The following graph (Figure 2) represents an example of determining the experimental standard uncertainty associated with the repeatability of the balance measurements (GUM: type A evaluation) for testing volume 10 µl. Figure2. Evaluation of type A input quantities Lab World Magazine 18 Vol. 03 No. 03 Feb-April 2014 If an outlying value can be traced back to a pipetting error, then it is permissible to repeat the measurement in question. If such a retrace does not come up with any failure, then the outlying value should be considered as a real but rare incident and kept in the data set. 2.2. Analysis of the measurement model and calculation of the measurement uncertainty According to the GUM Supplement 1 [6] several conditions must be fulfilled for the valid implementation of the GUM Uncertainty Framework in the process of the measurement uncertainty calculation. Figure 3 presents the results of the measurement model analysis, which includes the following tests and calculations: ● Linearity test for each input quantity in sixth areas of the distribution interval ● Calculation and validation of the results of the equivalent linear model ● Determination of the symmetry and the distribution type of the result quantities ● Check for correlated input quantities with a finite degree of freedom ● Check for non-linear correlated or non-normally distributed input quantities Figure3.Expert analysis of the measurement model The linearity of the model and the normal distribution of the result quantity are confirmed by conducting several analyses of the measurement model with different input. The recommended GUF method, corresponding to the standard GUM Uncertainty Framework, is applied for calculation of the measurement uncertainty. The QMSys GUM software calculates the partial derivatives (the first term of a Taylor series) to determine the sensitivity coefficients of the equivalent linear model and then calculates the combined standard uncertainty in accordance with the Gaussian error propagation law. The value of the coverage factor is determined automatically to the selected distribution and the specified coverage probability. 2.3. Measurement uncertainty budget The result of the analysis is presented as a clearly structured measurement uncertainty budget (Fig. 4) in a table form. This table holds all model quantities with their short names, value, associated standard uncertainty, distribution type, degrees of freedom, sensitivity coefficient and contribution to the combined standard uncertainty of the output quantity. Interim results are shown with the value and the standard uncertainty. Vol. 03 No. 03 Feb-April 2014 19 Lab World Magazine Figure4.Measurement uncertainty budget (GUF method) The result quantity is displayed in the bottom line with its value, corresponding combined standard uncertainty and the effective degrees of freedom. Finally, the complete result of the examination is presented as a value with associated expanded uncertainty, coverage factor, coverage probability and distribution type. It should be kept in mind that some of the numerical values of the sensitivity coefficients are volume dependent; therefore, it is not possible to use the values given in the example for other volumes. When the uncertainty calculation by the Monte-Carlo method is activated, the software automatically validates the results of the GUF Method by comparing the values, the combined standard uncertainties and the limits of the coverage intervals. The Monte Carlo method (Fig. 5) displays a histogram, statistical parameters of the estimated distribution of the result quantities and validation of the results. The result of the uncertainty analysis together with all input data can be printed with the help of configurable templates in MS Word or Excel format. Each analysis can be completely saved in a file with a selectable name. In this way, the examination is available at any time for a later review or editing. Each saved analysis can be used as a starting point for new uncertainty analyses using the same model, but with new or changed data sets. 3. Automated Uncertainty Calculation for Several Measurement Seriesin MS Excel In this section, an additional application of the developed model for calculation of measurement uncertainty is suggested, namely, with an Excell Add-In program. The full integration of calculating the measurement uncertainties in MS Excel is implemented by using the software QMSys GUMX (Excel Add-In). This program also enables calculation of measurement uncertainty in MS Excel for unlimited number of measurement series (result quantities with identical measurement model), using a model file for only one set of measurements. This functionality is particularly useful when calibrating in several points of the measuring range. In addition, it simplifies the modelling of the measurement process. The software QMSys GUMX reads the parameters of the input quantities from the current Excel file, calculates the measurement uncertainty and then exports the results in the same Excel file. The cells with the parameters of the input quantities are assigned in the model file with cell array names, defined in the Excel file. The parameters of the result quantities and the statistical evaluation type A input quantities can be exported by columns or rows. The following steps represent the procedure for calculating the uncertainty in MS Excel: Lab World Magazine 20 Vol. 03 No. 03 Feb-April 2014 ● Development of Excel files containing the data to be imported to the input quantities and coded cells for the export of the results of the uncertainty calculation ● Development of model file for calculating the measurement uncertainty with the QMSys GUM software, selecting the appropriate cells or cell ranges of the input quantities in the model file. ● Assigning of the model file to the Excel file; up to three different model files can be selected in one Excel file On the Figure 6 are shown the worksheets for entering the calibration data, calculating the measurement results and generating the calibration certificate. Figure5.Measurement uncertainty budget (Monte-Carlo method) The blue cells indicate entries of the volume measurement series and the environmental conditions. In the yellow cells are calculated the volume-dependent parameters of the additional input quantities. Following the described procedure, Excel files for calibrating multi-channel pipettes that are using the same model file for calculating the measurement uncertainty are also developed for each tested volume and channel. 4. Summary The developed measurement model for pipette calibration by the gravimetric method includes all relevant factors that influence the calibration results. The linearity of the model and the normal distribution of the result quantity are confirmed by conducting the model analysis according to the GUM Supplement 1. The significant uncertainty contributions, causing over 99% of the combined standard uncertainty of the pipette calibration, are: ● Repeatability of the balance measurements of the delivered volume ● Calibration of the balance ● Temperature difference water - pipette – air ● Operator contributions ● Rounding error due to balance finite resolution Furthermore, excluding the balance resolution, all significant uncertainty contributions depend on the measured volume. The repeatability of the balance measurements is specific for each channel of the multi-channel pipettes. Therefore, when calibrating piston pipettes by the gravimetric method, the measurement uncertainty should be calculated separately for every measured volume and channel. Vol. 03 No. 03 Feb-April 2014 21 Lab World Magazine Finally, the article acknowledges the complexity of measurement uncertainty analysis and the importance of valid results and reliable interpretations, thus, suggests two practical solutions by means of a professional software product. Figure 6. Calibration worksheet in MS Excel Lab World Magazine 22 Vol. 03 No. 03 Feb-April 2014 Figure 7 . Calibration certificate in MS Excel Vol. 03 No. 03 Feb-April 2014 23 Lab World Magazine About the Software: References [1] [2] [3] [4] [5] [6] [7] ISO 8655-6:2002 Piston-operated volumetric apparatus - Pt 6: Gravimetric methods for the determination of measurement error. Guideline DKD-R 8-1 Kalibrierung von Kolbenhubpipetten mit Luftpolster(12/2011) Blues, J., Bayliss, D.J., Buckley, M.: The calibration and use of piston pipettes, Measurement Good Practice Guide No. 69(07/2004) ISO/TR 20461:2000Determination of uncertainty for volume measurements made using the gravimetric method JCGM 100:2008 Evaluation of measurement data – Guide to the expression ofuncertainty in measurement JCGM 101:2008 Evaluation of measurement data Supplement 1 to the “GUM” - Propagation of distributions using a Monte Carlo method EURAMET/cg-19 Guidelines on the determination of uncertainty in gravimetric vol. calibration, Ver. 2.1 (03/2012) QMSys GUM software is a comprehensive tool for analysis of the measurement uncertainty of physical measurements, chemical analyses and calibrations. The program supports the systematic procedure in building an uncertainty analysis, as requested in the corresponding standards and guides, and reduces significantly the analytical and computational effort. The user of the software needs to furnish only the formulation stage, while the program performs automatically all other stages of the measurement uncertainty analysis. The software allows the user to freely enter or modify the model equation. With this feature, the application can be used to evaluate measurement uncertainty of almost any measurement process. For further information, please visit: www.qsyst.com. The Author About the Author: Stefan Golemanov graduated with a Master of Science in Mechanical and Precision Engineering (M.Sc.Eng.) at the Technical University of Gabrovo, Bulgaria. After performing specializations in Austria and Germany focused on metrology and quality management, he founded in 1995 with partners the company Qualisyst Ltd., which is specialized in the development of metrology, quality assu rance and quality management software. Contact Email: Stefan@qsyst.com Advertisement Vol. 03 No. 03 Feb-April 2014 24 Lab World Magazine