atypiques test de conformité de produits fileEveryone involved with food/feed, from the farmer to...

Détection de spectres anormaux \ atypiques test de conformité de produitsJuan Antonio Fernández PiernaVincent Baeten & Pierre Dardenne*Walloon Agricultural Research Centre (CRA-W)Gembloux, Belgium* Ex CRA-W member

j.fernandez@cra.wallonie.bev.baeten@cra.wallonie.be

EXPERTISES

Calibration Instrument network

ISO17025&

ISO17043

Interlaboratorystudy

Training

Spectral band interpretation

Homogeneitystudy

Chemometrics

Sampling

Samplepresentation

Standardisation

Data bases

Food and Feed Quality Unit

Ensuring an acceptable level of food/feed quality and safety isnecessary to provide adequate protection for consumers and tofacilitate trade.

This can be achieved by implementing and monitoring qualityassurance measures along the entire food/feed chain - from thereceipt of the raw materials (primary products and otheringredients) to the shipping and marketing of the final productsto the consumers - when it is appropriate and when it is possible.Everyone involved with food/feed, from the farmer to theconsumer, shares in the responsibilities to keep the food/feedsupply safe by taking the necessary precautions to keep food/feedprotected from hazards that can increase human/animal health risks.

The contextQuality control and assurance

In developing countries, the Agrofood/feedindustry is in the process of rapid integration.Medium and large scale groups are expandingproduction capacities

Food and feed industries need todistinguish themselves by suppling afinal quality product. For this, qualitycontrol analysis should be performed,not only in the final product but,mainly at the entrance of theproduction chain when the rawmaterial reaches the industry.

More serious competition!

The contextIndustry

Spectroscopy

Ana

lytic

al M

etho

d

Process control

The contextNeed of rapid methods

The contextVibrational Spectroscopy

1200 1400 1600 1800 2000 2200 24000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

wavelength (nm)

log

(1/R

)

soyabean mealFull fat soyaSoya hulls

Regression equations

Local Window

PCA

GHsimilarity

Several criteria

Protein

Fat

The aimNIR analysis & chemometrics procedure

PLS models

Protein

Fat

Regression equations

ProcedureNIR analysis & chemometrics

Soybean meal models from

Haaland & Thomas, 1988

The Global H (GH) criteria is a modification of the Mahalanobisdistance of each sample from the average spectrum. This distancetakes into account the correlation of the data set.

GH

Mahalanobis

Euclidean

ProcedureNIR analysis & chemometrics

GH value < 3 will guaranteehomogeneous spectra

GH value > 3 will guarantee thatthe spectra are a result of randomchance and hence are most likelyto have outliers

Local Window Principal Component Analysis (LWPCA)Local Window

PCA

ProcedureNIR analysis & chemometrics

ProcedureLocal discrimination modelling

Residuals are used to build thresholds to be applied to theresiduals of the samples to be predicted.

Residual plots

ProcedureNIR analysis & chemometrics

Study case 1Feed contamination

Contamination by microorganism and animal by-products (salmonella, animal proteins…)

Contamination by persistent organic pollutants and toxicmetals (melamine, dioxins, brominated flame retardants…)

Natural toxins in animal feed (mycotoxins, plant toxins…)

Veterinary medicinal products in feeds (antibiotics, …)

Risks from emerging technologies (GM crops, biofuel and food industry by-products…)

Study case 1Feed contamination

Study case 1Feed contamination

The aim of this work is to propose a procedure based on NearInfrared (NIR) spectroscopy and Chemometrics in order tocharacterize a typical feed product (soybean meal) as well as todetect the presence of any possible contaminant.

Laboratory level• Melamine• Whey

Feed plant level• Whey

Study case 1Feed contamination

NIR/NIR Hyperspectral imaging and chemometricsmake it possible to compare efficiently suspectedspectra of new samples and detect melamine in feed.

Melamine

Soybean meal non contaminated

Soybean meal contaminated with 5%

melamine

Rapid detection of any modification

by non-authorized additives likeMelamine

Study case 1Melamine in soybean meal

No false positives

Lab level - whey

No false positives

Study case 1Feed contamination

Lab level - whey

Study case 1Feed contamination

Extension to feed millsStudy case 1Feed contamination

STEP 1Loading of 5 Tons

of Soya bean directly from the

truck

STEP 2Loading of 5 Tons

of Soya bean from the truck

contaminated by simulating a local

contamination

STEP 3Loading of 5 Tons

of Soya bean directly from the

truck

STEP 4Loading of 5 Tons

of Soya bean from the truck

contaminated by mixing

STEP 5Emptying of the

truck. No deliberate

pollution done

Feed mill contamination

Study case 1Feed contamination

Protein Fat

GH

Whey – set 1

Study case 1Feed contamination

Feed mill contamination

LWPCA

Feed mill contamination

Study case 1Feed contamination

Study case 2Milk contamination

Study case 2Milk contamination

Study case 2Milk contamination

These results show that no clear conclusion can be obtainedwhen looking directly at the spectra.

GH values detect abnormalities at levels higher than 500 ppm.

LWPCA allows detecting contamination at levels up to 100ppm; however at those levels the detection of melamine in milkbecomes unstable, which is an indication that the technique hasprobably reached its limit of detection.

Study case 2Milk contamination

Local moving window PCA method is proposed for thecharacterization of agronomical products and the detection ofpossible contaminants using vibrational spectroscopy.

In the examples presented here, soybean meal and liquid UHTsamples have been contaminated with melamine/whey, makingthem thus semi-targeted studies.

However the method should be used as a method for the untargeteddetection of abnormalities (real contamination or fraud) in the dataand a previous step for further analyses.

Local discrimination modelling

Conclusion

NIR spectroscopy in combination with a chemometricsbased protocol is a perfect tool, at laboratory and atindustry levels, for:

- the characterization of food/feed materials- Detection of targeted and untargeted

adulteration / contamination

Conclusion

http://www.cra.wallonie.be/en/events/

More info?

Thanks!