Modeling Different Products for NIR Transmitters – Global Vs. Separate Product Calibration

Global vs. Local Modeling for Different Products with NIR Transmitters

Near IR transmitters can be calibrated with separate channels for each product or they can utilize a global calibration.  A Global calibration is easier to generate while separate calibrations for each product usually provides tighter standard errors.

Illustrated below are three different product results based on Global versus individual (local) product codes (calibrations) within a product family.  The application data is for the measurement of PSA coat weight in the range of 8 to 40 GSM (grams per square meter) on polyethylene (PE) film at 25, 35, and 55 mils thickness.  The chart and graphs below compare results of a local calibration for each coat weight measurement on film with various thickness versus the PSA measurement of all with a single global calibration and the impact it has on accuracy and standard error.

Film Thickness

MCT PSA Accuracy

Local Model (+/-) GSM

MCT PSA Accuracy Global Model (+/-) GSM

MCT PSA Standard Error Local Model (+/-) GSM

MCT PSA Standard Error Global Model (+/-) GSM

Correlation

Coefficient

25 mils 0.1 0.3 0.05 0.05 0.99
35 mils 0.1 0.6 0.1 0.2 0.99
55 mils 0.1 0.4 0.1 0.4 0.98
All 1.0 0.43 0.99

 

Accuracy:  How close measurement is to target.  E.g. how close a dart is to the bull’s eye.  Accuracy is calculated from the Standard Deviation of Calibration Residual Values (difference between Lab and Calibrated MCT Values).

Precision:  How close the subsequent dart is to initial dart for same conditions (reproducibility/repeatability).

Standard Error:  Standard deviation of sample distribution for sample population mean or standard error of mean.  Think standard deviation and variable depending on sample population.

Standard Deviation:  How much variation from the average value.  Low SD indicates data points close to mean and high SD indicates values spread out over large range of values.

Correlation Coefficient:  Linear relationship between two variables, in this case Lab vs. MCT PSA mils thickness measurement.  All Correlation Coefficients (CC) are 0.99 which is close to perfect at 1.0.  Non-linear relationships can also be handled with NIR software technology, though not necessary in these applications.  Low CC indicates a low performance expectation.

 

PSA coating on PE

Universal Calibration Across Product Changes

 

 

NIR Transmitters Calibration Approach Comparison

NIR Calibrations for Single Products Vs. Global Calibration

 

Speak Your Mind

*