Application Note

AN 3486
Rev. 2

NIRS™ DS2500 / DS2500 F

Compound Feed

NIRS DS2500 / DS2500 F provides fast and

accurate results for compound feed parameters.
It allows you to perform frequent checks against
production specifications and control the quality
of the product.
By using a FOSS NIR solution for compound
feed analysis you will get results almost
instantly instead of waiting hours for results
from standard wet chemistry analysis in the
laboratory.
This Application Note describes the calibration
models and the results that can be expected
when analysing ground samples of compound
feed.

Samples Used in Calibration Set

The application is developed with following feed sample:
• Cattle • Poultry
• Calf cattle • Broiler starter
• Milk cattle • Broiler finisher
• Pig • Broiler grower
• Piglet • Chick feed
• Pig sow feed • Layer Feed 1)

1)
Layer Feed often has an addition of limestone/calcium to it which affects the ash concentration and there is
therefore a specific ash calibration for this sample type.

Application Note 3486 / Rev. 2 Part No. 6007 2931 1(6)

Calibration Information
The application covers the following parameters and ranges:

Parameter Model PM version Moisture Basis N Min Max

1)
Protein ANN 2.0.0.0 As Is 7145 1.61 51

Moisture ANN 2.0.0.0 n/a 4125 6.45 14.9

Fat ANN 2.0.0.0 As Is 4415 0.7 12.59

Fibre ANN 2.0.0.0 As Is 3518 0.95 18.9

Starch ANN 2.0.0.0 As Is 3453 1.64 58.4

Ash I ANN 2.0.0.0 As Is 3496 3.2 9.98

Ash II (Layer Feed) PLS 2.0.0.0 As Is 662 3.69 17.49

1)
The Nitrogen conversion factor for Feed applications is 6.25
Model: ANN (Artificial Neural Network); PLS (Partial Least Square)
PM version: Prediction model version number
Moisture Basis: As Is = no moisture compensation; not applicable for the moisture parameter
N: Number of individual samples in the calibration set
Min: Minimum reference value in the calibration set
Max: Maximum reference value in the calibration set

Performance
Performance was evaluated using independent validation test sets which are not included in the
calibration set representing data from different instruments and a different reference laboratories.

Total Validation Test Set

Mix of Feed samples (ground and unground)
Parameter N Acc. Min Max RSQ

Protein 1277 0.92 9.53 32.4 0.88

Moisture 770 0.33 8.3 14.5 0.89

Fat 1183 0.53 0.7 10.7 0.89

Fibre 974 0.69 0.64 14.62 0.86

Starch 444 1.56 2.5 52.23 0.93

Ash I 574 0.53 3.38 8.43 0.72

Ash II (Layer Feed) 63 1.79 4.55 17.76 0.7

N: Number of individual samples in the validation set

Acc. Accuracy expressed as Standard Error of Prediction (SEP) corrected for bias.
For Layer Feed it is expressed as Standard Error of Cross Validation (SECV).
Min: Minimum reference value in the validation set
Max: Maximum reference value in the validation set
RSQ: Linear correlation between predicted results and reference results.

Application Note 3486 / Rev. 2 Part No. 6007 2931 2(6)

Validation Data for Specific Sub-classes
Poultry Feed, ground
Parameter N Acc. Min Max RSQ

Protein 594 1.03 9.53 23.81 0.81

Moisture 313 0.31 8.3 13.06 0.86

Fat 566 0.53 1 10.7 0.86

Fibre 454 0.53 1.44 6.88 0.61

Starch 198 1.39 28.18 47.5 0.87

Ash 306 0.55 4.4 8.36 0.61

Pig Feed, unground

Parameter N Acc. Min Max RSQ

Protein 512 0.63 10.8 32.4 0.9

Moisture 369 0.32 8.35 13.8 0.89

Fat 385 0.43 0.7 9.15 0.94

Fibre 443 0.57 0.64 9.8 0.87

Starch 184 1.43 2.5 51.75 0.92

Ash 229 0.48 3.38 8.43 0.81

Cattle Feed, unground

Parameter N Acc. Min Max RSQ

Protein 137 1.1 15.03 25.39 0.85

Moisture 56 0.43 9.5 14.5 0.82

Fat 199 0.55 1.32 6.94 0.85

Fibre 44 1.25 2.4 14.62 0.71

Starch 30 1.46 19.6 34.2 0.87

Application Note 3486 / Rev. 2 Part No. 6007 2931 3(6)

Note:
Graphics showing predicted results vs. reference analysis for the total
validation set.

Total Validation Set

Protein Moisture

Fat Fibre

Starch Ash

Application Note 3486 / Rev. 2 Part No. 6007 2931 4(6)

Note:
The performance example outlined in this note should only be regarded as a
guideline for the expected performance of new installations. The
performance will always depend on the uniformity of the sample
preparation and the homogeneity of the product, as well as the accuracy of
the reference method used and the range for the test samples. An indication
of the obtainable performance can be found as approximately 1.5 to 2 times
the reproducibility of the reference method.
If the samples you are measuring exceed the stated calibration ranges, or have non-common
variations of other components, this might also influence the performance of the calibrations. For
each unknown sample analyzed the spectra of it will be compared to the samples in the calibration
database. Two key values will be given as an indicator to how well the unknown sample fit the
calibration samples:
• Global H value (GH) - measures how far the sample is from the center of the database
• Neighbourhood H value (NH) - measures how close the sample is to the nearest sample in the
database
(Default Warning and Action limits for these NH / GH values are set for each prediction model
in the software.)

Sampling, Sample Preparation, Sample Cup and Analysis

If grinding of the product is required, we recommend a cyclone mill such as Retch or Cyclotec,
with a 1 mm sieve (see “Note on grinding” below).
The recommended way to analyse ground samples with NIRS DS2500/DS2500 F is to use the
standard Small cup and for unground samples the recommendation is to use the Large cup.
Sample volumes (volumes will vary with sample density):
• Small cup: 30 grams
• Large cup: 200 grams
Compressing the sample is not necessary.
We recommend that the analyser lid is closed during analysis.

Note on Grinding
Homogenous samples do generally not require grinding. Inhomogeneous samples can be measured
unground but performance and repeatability will benefit from grinding. For some seeds where the
hulls do not allow enough light to penetrate the recommendation is to always grind the sample.
If in doubt, scan the unground sample 10 times, refilling the sample cup between each scan.
Calculate the standard deviation across the 10 repacks. If the standard deviation is larger than the
accuracy presented above, grinding is recommended.

Sample Temperature
The calibrations have been stabilised to be able to analyse samples at sample temperatures between
10-30 ºC.
For optimum performance, the sample temperature should be as close as possible to the ambient
temperature. Analysing at a temperature higher or lower than the environment temperature
increases the risk for moisture condensation or moisture evaporation from the sample and
temperature drift during the analysis.

Application Note 3486 / Rev. 2 Part No. 6007 2931 5(6)

Reference Methods
• Fat ISO 6492:1999 Animal feeding stuffs -
Determination of fat content
• Moisture ISO 6496:1999 Animal Feeding Stuffs -
Determination of moisture and other volatile matter content
• Crude Protein ISO 5983:1997 Animal feeding stuffs -
Determination of nitrogen content and calculation of crude protein content -
Kjeldahl method
• Crude Fibre ISO 6865:2000 Animal feeding stuffs -
Determination of crude fibre content - Method with intermediate filtration
• Crude Ash ISO 5984:2002 Animal feeding stuffs -
Determination of crude ash
• Starch ISO 6493:2000 Animal feeding stuffs -
Determination of starch content - Polarimetric method

Ordering Information
• 60071850 Compound Feed - NIRS DS2500 / DS2500 F
• 60072798 Ash for Layer Feed - NIRS DS2500 / DS2500 F

FOSS Analytical A/S Tel +45 7010 3370 FOSS Analytical Co., Ltd. Tel +86 512 6292 0100
Foss Allé 1 Fax +45 7010 3371 6 Louyang Road, Building 1 Fax +86 512 6280 5630
DK-3400 Hillerød E-mail info@foss.dk 215121, SIP, Suzhou E-mail info@foss.dk
Denmark Web www.foss.dk P.R. China Web www.foss.dk

Application Note 3486 / Rev. 2 Part No. 6007 2931 6(6)