10 / 07 BRUINS INSTRUMENTS 1 BRUINS INSTRUMENTS BRUINS INSTRUMENTS Training Near Infrared Transmission Whole Grain Analyzer
Dec 26, 2015
10 / 07 BRUINS INSTRUMENTS 1
BRUINS INSTRUMENTSBRUINS INSTRUMENTS
Training
Near Infrared Transmission
Whole Grain Analyzer
10 / 07BRUINS INSTRUMENTS 2
Targets:Targets:
History and Basics of NIRCalibration developmentToubleshooting & Maintainance
10 / 07BRUINS INSTRUMENTS 3
History of NIRHistory of NIR
First detected by William Herschel beginning of 19th century
First instrument built by W.W. Coblentz about 1900 First practical use in the 1930s First agriculture usage by Karl Norris 1968 First commercial scanning instruments 1978 First BRUINS NIR spectrophotometer 1979 First BRUINS NIT spectrophotometer 1982
10 / 07BRUINS INSTRUMENTS 4
Basics of NIRBasics of NIR
NIR = NIR = NNear ear IInfranfraRRed Spectroscopyed Spectroscopy
Wavelength of Light
UV > VIS > NIR>(M)IR
nm 380 750 / 11002500
(NIT / NIR)
Energy of Light
10 / 07BRUINS INSTRUMENTS 5
Basics of NIRBasics of NIR
Typical properties analysed by NIR are
organic compounds like
Oil
Protein
Starch
And
Water
Useful concentration are in % range,
ppm or ppb level are normally impossible.
10 / 07BRUINS INSTRUMENTS 6
Basics of NIRBasics of NIR
NIR active groups
O-H Water, Alcohol
C-H Carbohydrates (Starch, Sugar, Cellulose)
Fat / Oil
N-H Protein
10 / 07BRUINS INSTRUMENTS 7
Basics of NIR Basics of NIR
Molekular Vibrations
O
H H
10 / 07BRUINS INSTRUMENTS 8
Basics of NIR Basics of NIR
Molekular Vibrations
10 / 07BRUINS INSTRUMENTS 9
Basics of NIR Basics of NIR
Molekular Vibrations
10 / 07BRUINS INSTRUMENTS 10
CalibrationCalibration
Reference Analysis Sample selection Sampling Sample Preparation / Homogenisation Sample Presentation
10 / 07BRUINS INSTRUMENTS 11
Calibration: Reference AnalysisCalibration: Reference Analysis
NIR calibration will be dependend from reference analysis, so
Select best possible method Investigate „real“ method error Make sure that the same sample material will be
analysed Use at least double analysis Comment all specialities of actual sample
10 / 07BRUINS INSTRUMENTS 12
Calibration: Lab Error InfluenceCalibration: Lab Error Influence
10 / 07BRUINS INSTRUMENTS 13
Calibration: Lab Error InfluenceCalibration: Lab Error Influence
10 / 07BRUINS INSTRUMENTS 14
Calibration: Sample SelectionCalibration: Sample Selection
The complete interesting concentration range must be covered by calibration samples.
The calibration set must also include samples with changes from other constituents, eg. low protein with high moisture, low protein with low moisture etc.
Further samples must be included which are representing influences from climate, soil, cultivars etc.
Due to the effect of temperature to NIR also cold and warm samples should be added.
10 / 07BRUINS INSTRUMENTS 15
Calibration: SamplingCalibration: Sampling
The sample must be representative for the total amount.
Take subsamples from different locations. Thoroughly mix these to get a homogene sample.
10 / 07BRUINS INSTRUMENTS 16
Calibration: Sample PreparationCalibration: Sample Preparation
No extra preparation needed for most grain and oil seeds.
Samples with thick shell must be ground (e.g. sunflower seeds).
The same sample preparation must be used for calibration and routine analysis.
By grinding a sample the moisture content will decrease.
Select the best grinder to reduce this moisture loss.
If a grinder with a sieve will be used, control the size of the mesh.
10 / 07BRUINS INSTRUMENTS 17
Calibration: Sample PresentationCalibration: Sample Presentation
OmegAnalyzer and AgriCheck uses the NIR transmission, so for such meal the MultiCheck will be the better choice
As an alternative, but less efficient, a micro sample cup for meal or powders can be used
10 / 07BRUINS INSTRUMENTS 18
Calibration: Sample PresentationCalibration: Sample Presentation
In all cases, if fine powder or ground sample: Thoroughly mix the sample before filling In cases of components with different density
make sure, that no material separate
If a micro sample cup is used: Make sure that the sample is filled without wholes Knock a bit, that the sample will slide together,
but not to often because different components may separate
10 / 07BRUINS INSTRUMENTS 19
Calibration: StandardisationCalibration: Standardisation
Before using the instrument in routine, it must be compared against the reference method with a set of samples.
These samples should cover the range of concentrations and normal variations in moisture, area and genotypes.
In order to get a good adjustment, the sample must be well mixed that the same material will be analysed with both methods.
The data for the reference analysis need to be well controlled, at least a double analysis must be performed.
10 / 07BRUINS INSTRUMENTS 20
Calibration: StandardisationCalibration: Standardisation
The easiest comparison of accuracy and necessary adjustment can be made with Excel.
Just two columns need to filled , one with reference analysis the second with predicted NIR data.
Using the Excel functions for mean, standard error xy and correlation will calculate all necessary information.
Additional the Excel graph can be used to show how good the predicted values fit to the reference analysis.
10 / 07BRUINS INSTRUMENTS 21
Calibration: BiasCalibration: Bias
Actual Predicted8
9
10
11
12
13
14
15
16
17
18
7.5
8.5
9.7
10.6
11.7
12.4
13.7
14.4
15.5
16.6
17.7
10 / 07BRUINS INSTRUMENTS 22
Calibration: BiasCalibration: Bias
Actual Predicted
8
9
10
11
12
13
14
15
16
17
18
7.5
8.5
9.7
10.6
11.7
12.4
13.7
14.4
15.5
16.6
17.7
Actual Predicted
Mean 13.00 12.57
SEP 0.121
R² 0.999
Bias - 0.43
Slope 1.001
10 / 07BRUINS INSTRUMENTS 23
Calibration: BiasCalibration: Bias
10 / 07BRUINS INSTRUMENTS 24
Calibration: Bias, 0.43 addedCalibration: Bias, 0.43 added
Actual Predicted8
9
10
11
12
13
14
15
16
17
18
7.93
8.93
10.13
11.03
12.13
12.83
14.13
14.83
15.93
17.03
18.08
10 / 07BRUINS INSTRUMENTS 25
Calibration: BiasCalibration: Bias
10 / 07BRUINS INSTRUMENTS 26
Calibration: SlopeCalibration: Slope
While the bias will move all data parallel, the slope will change the extreme values more than the values around the middle.
So adjusting the slope will much more influence the predicted result.
For the slope adjustment the sample set must really cover the full calibration range.
Normally this adjustment should not be used. If it seems to be necessary, look for the reason. It might be better to select or built another calibration.
10 / 07BRUINS INSTRUMENTS 27
Calibration: SlopeCalibration: Slope
Actual Predicted8
9
10
11
12
13
14
15
16
17
18
7.5
8.3
9.8
10.5
12.1
12.9
14.5
15.0
16.6
17.2
18.5
10 / 07BRUINS INSTRUMENTS 28
Calibration: SlopeCalibration: Slope
10 / 07BRUINS INSTRUMENTS 29
Calibration: corrected SlopeCalibration: corrected Slope
Actual Predicted8
9
10
11
12
13
14
15
16
17
18
8.14
8.86
10.21
10.83
12.27
12.99
14.43
14.88
16.31
16.85
18.02
10 / 07BRUINS INSTRUMENTS 30
Calibration: corrected SlopeCalibration: corrected Slope
10 / 07BRUINS INSTRUMENTS 31
Calibration: SlopeCalibration: Slope
Actual Predicted11
12
13
14
10.5
12.1
12.9
14.5
10 / 07BRUINS INSTRUMENTS 32
Calibration: SlopeCalibration: Slope
10 / 07BRUINS INSTRUMENTS 33
Calibration: SlopeCalibration: Slope
10 / 07BRUINS INSTRUMENTS 34
Calibration: corrected Slope?Calibration: corrected Slope?
Actual Predicted8
9
10
11
12
13
14
15
16
17
18
8.59
9.21
10.39
10.93
12.18
12.81
14.06
14.45
15.70
16.17
17.18
10 / 07BRUINS INSTRUMENTS 35
Calibration: corrected Slope?Calibration: corrected Slope?