quality measurements using nir/mir spectroscopy: a rotten apple could turn your product into a lemon
DESCRIPTION
Light interacts with a product's organic molecules causing variations in light absorption. The transmitted or reflected light can be measured with a spectrometer and the resultant spectral signature used to qualify or quantify properties of the product. The discussion will include - how light interacts with molecules, characteristics of the different electromagnetic spectral bands, in-line hardware required to collect light, and fundamentals of chemometrics. Presenter -- Gary Brown Gary Brown is one of the principle engineers with Australian Innovative Engineering and has spent the last 12+ years developing in-line instrumentation using NIR spectroscopy to measure properties of fresh fruit. He is now concentrating his efforts in applying the technology for in-line product authentication for the food and pharmaceutical industries.TRANSCRIPT
A rotten apple could turn your product into a Lemon
NIR/MIR Spectroscopy
Focusing on Inline Use
The Spectrum
Region Wavelength Wavenumbers (cm-1) Frequencies (Hz)
Near UV 200 to 350 nm
Vis 400 to 700 nm Xx to 12800
NIR 700 to 2500 nm 12800 to 4000 3.8x1014 to 1.2x1014
Mid IR 2.5 to 50 um 4000 to 200
Far IR 50 to 100 um 200 to 10
How light is used
How light is used
Glucose Water Spectra
Regions
• Near UV – electronic transitions of the energetic levels of valence orbitals, absorption of peptidic bonds in proteins, and aromatic amino acids.
• Vis – electronic transitions occur in molecules with large numbers of conjugated double bonds. i.e. carotenoids, chlorophylls, and porphyrins.
Regions cont…..
• NIR – first spectral region exhibiting absorption bands related to molecule vibrations, widely used for composition analysis of food products.
• MIR – main region of vibrational spectroscopy. This region retains information allowing organic molecules such as proteins, polysaccharides, and lipids to be characterized.
Molecular spectroscopy
• Analysis and quantification of molecular responses to introduced radiation.
• Energy exchange occurs between the radiation energy and the energy contained within the molecule.
Molecular spectroscopy…………….
See Wiki for live example
Molecular Groups
• Sensitive to molecules containing C-H, O-H, and NH bonds.
• Interact with NIR portion of spectrum• Starch and sugars (C-H)• Alcohols, moisture and acids (O-H)• Protein (N-H)
MIR versus NIR
• MIR absorption fundamental vibration energies in the mid-IR part of the electromagnetic spectrum.
• NIR absorption overtones 1st, 2nd, and 3rd and combinations of CH, NH, and OH vibrations occur in the near-IR part of the electromagnetic spectrum.
NIR overtones and combinations
NIR region
Beers Law = € ℓ c
BEER-LAMBERT, LAMBERT-BEER, BEER-LAMBERT-BOUQUER.
Beers Law …………
Absorbance = € ℓ c
€ = molar absorptivity, ℓ = path length, and c = concentration.
The Beer’s not perfect
Deviates because• Particle scatter• Interferents, minute contaminants• Molecular interactions• Changes in refractive index• Stray light• Changes in sample size/path length.
Better Beer
Corrected by (data preprocessing)• MSC (multiplicative scatter correction)• SNV (standard normal variate correction) and
normalization• Baseline correction• Differentiation (Savitzky-Golay)
Collection of Spectra
Diffuse Reflection
Interactance
Beers Law …………again
Absorbance = log10 ( Iref/I )= € ℓ c
Iref = intensity of source light
I = intensity of light through sample
€ = molar absorptivity, ℓ = path length, and c = concentration.
How to measure
Wavelength Selection
Filters• Optical interference
(rotating disk with 9 x filter)• Optical Tunable Filters (AOTF)• Liquid crystal tunable filters (LCTF)
Wavelength Selection………..
Monochromator• Disperses light with a wide range of
wavelengths into monochromatic light at a different wavelength. (Rotating diffraction grating or interferometer).
• Classified as pre and post dispersive
Wavelength Selection………..Diffraction Grating
NIR region detectors
Material Range (cm-1) Wavelength Unit Wavelength UnitSilicon 16700-
9000599 nm 1111 nm
InGaAs (Indium Gallium Arsenide)
12000-6000
833 nm 1667 nm
PbSe (Lead Selenide) 11000-2000
909 nm 5000 nm
MCT (Mercury Cadmium Telluride)
117000-400
85 nm 25000 nm
DTGS (Deuterated triglycine sulfate)
12000-350 833 nm 28571 nm
MIR region detectors
Material Range (cm-1) Wavelength Unit Wavelength Unit
DTGS / KBr 12000-350 833 nm 29 umDTGS / Csl 6400-200 1563 nm 50 um
MCT 11700-400 855 nm 25 umPhotoacoustic 10000-400 1000 nm 25 um
DTGS – thermo capacitive device, inexpensive but have low sensitivity and slow in response.MCT – semiconductor where IR radiation causes changes in electron conduction. Faster and more sensitive (higher SNR) than DTGS, but need to be cooled and have narrow bandwidth.KBr – Potassium Bromide, Csl
Transmission of Light
Transmission of Light……………….
Dispersive Spectrometer
Grating-based dispersive spectrometer• Low Cost ($1500 to $15k)• Fast spectrum acquisition (<10msec)• Silicone or InGaAs ( 300nm to 2.5um)• External trigger for inline use• Miniature models available• Hand Held units available• Electronic cooling down to -10 deg C.
Dispersive Spectrometer
FT-NIRPreferred method for food composition and
quality, why –• Quick spectrum acquisition (< 0.5 sec)• User friendly easy to use Chemometrics
packages• Inline usable via fiber optics• Stable and repeatable results• Superior sensitivity
FT-NIR…………….
FT-NIR…………….
FT compared to Dispersive
MIR pros/consMIR advantages are –• Part of the spectrum that contains fundamental vibrations• Well defined bands for organic functional groups• Good for qualitative and quantitative identification of
functional groupsMIR disadvantages -• Available energy drops off rapidly with increasing
wavelength• Expensive transmitting materials• High absorption means path lengths have to be small.• Sample preparation required
NIR prosNIR advantages are –• Cheap transmission with glass optics• Instruments simpler and cheaper to manufacture.• Non destructive, no sample preparation required because NIR
bands 10-100 times less intense.• Good for qualitative and quantitative identification via
combination bands and overtones of functional groups.• Weak absorption due to water overtones enables analysis of
high moisture products.• Lower absorption means longer path lengths. 1 to 10mm.• Extremely high signal-to-noise in spectral data enables
Chemometrics to extract compositional information.• Not influenced by CO2 eliminating instrument purging.
NIR consNIR disadvantages are -• Less information contained in spectra.• Combination and overtone bands make
association with individual chemical groups more difficult.
• Generally can’t indentify components of less than 1% in product.
• Need more robust calibration techniques.• Relies on Chemometrics – PCA, PLS, SIMCA• Robustness of calibrations needs to be monitored.
NIR example
Diffused Reflectance absorbance raw, and 2nd Derivative of Bacillus cereus.
MIR example
System Overview
ChemometricsThe practice of applying mathematical tools in
order to extract chemical or physical information from a dataset (NIR spectra).
Normally involves the following steps –• Data preprocessing (base line removal,
filtering, scatter correction)• Data reduction and visualisation. (PCA, SIMCA)• Outlier detection• Qualitative and Quantitative model
development. (PCA-R, PLS)
Model Development
Model Development
Unsupervised –• sample clusters in a multidimensional space
created by a Principle Components Regression (PCA-R)
• resulting model will predict group classification
• Samples which do not belong to a group can be classified as outliers
PCA-R
Austria( ), Switzerland(□), Germany( ), France Thermized(■ ▲ x), France Raw(o), and Finland(• ).
Model Development…………
Supervised –• means we have allocated a result for each
sample and a Partial Least Squares (PLS) regression generates a model to predict the result in future samples.
• Samples which do not belong to the model can be classified as outliers.
System Overview
Models in Prediction
Water Soluble Nitrogen (WSN) validation of FT-MIR recorded on European Emmental Cheeses produced during summer.
Non protein nitrogen (NPN) validation of FT-MIR recorded on European Emmental Cheeses produced during summer.
Data Reduction via PCA
For a photo diode array there will be 255 variables for each spectra. This is normally reduced down to less than 10 using PCA.
After mean centering.
PCA example
Defining our Results
RMSEP (Root Mean Square Error of Prediction)• RMSEP = SQRT(∑(ai-pi-bias)2/n-1)
where ai=actual value and pi=predicted value.
RMSECV (Root Mean Square Error of Cross Validation)
• Calculated as per RMSEP• Predicted results are determined for samples not
included in the initial calibration model. • Best indication of how well your model is doing.
Defining our Results…………………
R or R2 (coefficient of determination)• Quantifies how well the predicted-v-actual
values fit onto a straight line.
RPD• RPD = SD/SECV where SD=standard deviation.• RPD best if >3 for the model to be reliable.
NIR spectroscopy Advantages• Minimal to no sample preparation• Deeper sample penetration than Mid Nir• Able to measure many constituents simultaneously• High Scan Speed ( < 1sec)• High Resolution ( Grating – 0.2cm-1, FT – 0.1 to 0.005-1 )• Wide range of application ( almost all organic and some
inorganic )• Quantitative and Qualitative results• No phase constraints – gas, liquid or solid.• Non Destructive, non contact.• Faster, safer working environment that does not require
chemicals
NIR spectroscopy Disadvantages
• Some insight required in sample selection for model development.
• Black Box – not able to easily understand how results are determined.
• Model development and maintenance is an ongoing expense.
• Typically able to measure organic constituents above 1 % (approx)
Examples - Meat (Beef)• Packaged Beef - Fat, protein, water content in emulsified
meats. (1300-2000nm, R>0.9)• Online – Lean beef blended to increase %fat. Five wavelengths
to measure fat and water then calculated protein. (Wavelengths 1441, 1510, 1655, 1728, 1810nm). SEP_fat(1.5% for 7 to 26%), SEP_water(1.3% for 58 to 78%), and SEP_protein(0.7% for 15 to 21%).
• Intramuscular fat – RMSEP(1.2% for 1 to 14%) using 1100 -2500nm and R2>0.98.
• Tenderness – SEP(1.2% for 1 to 9 classification) and R2=0.65. (needs work)
• Warner-Bratzler shear force (WBSF) – longissimus thoracis steaks > 79% correct classification. WBSF SEP(1.2kg for 2 to 11.7kg) with R2=0.67, RMSECV=1.3kg.
Examples - Meat (Beef)…………….• Cooking end-point temperature (EPT) – critical for safe
consumption of beef. Temp (high, long time) = not palatable. Temp (lower, shorter time) = increase food piosoning. EPS SEP(0.74degC ) with R2=0.97 using 400-2500nm.
• Beef adulteration with lamb, pork, skim milk powder, wheat flour.
• Distinguishing frozen-then-thawed then minced– 100% correct classification of frozen-thawed samples, 19% error for fresh samples.
• Microbial spoilage – PH influences microbial growth, 1413 and 1405 cm-1 identified as peaks indicative of amide-CN due to protein degradation by microorganisms.
Examples - Meat (Pork)• Intact sausages measuring fat, moisture and
protein – Fat SEP 1.47% with R2=0.98, moisture SEP 0.97% with R2=0.93, and protein SEP 1.08% with R2=0.97.
• Fatty Acid Composition • PH – SEP(0.1 for 5.3 to 6.7) with R=0.73,
RPD=0.25/0.1 = 2.5, 1000-2630nm.• Water Holding Capacity (WHC) – SEP(1.8% for
0.7 to 8%) with R=0.84. NIR reflectance able to correctly classify samples <5% or >7% WHC.
Examples - Meat (Chicken)• PH, colour, shear force, tough and tender
classification of cooked and raw meat using Vis/NIr.
• Fecal contamination on chicken skins using Vis/NIR
• Microbial spoilage, total viable counts using FTIR/Machine learning.
Examples - Dairy (Milk)• Inline during milking to predict fat, protein,
lactose, somatic cell count and milk urea nitrogen. Achieved R2 between 0.82 and 0.95 with standard errors between 0.05 and 1.33.
• Protein, fat, casein, whey protein, lactose, dry matter for raw milk.
• Fat, protein, dry matter for processed milk.
Examples - Dairy
Cheese• Dry Matter, Fat, moisture• Cholesterol (Paradkar et al 2002)Butter• Moisture, saltPowder • Water, fat, protein, lactose.
Examples
Margarine (inline)• Moisture RMSECV 0.3% (weight) with
R2=0.998 (780-1100nm)Honey• Adulteration with fructose and glucose. Pure
honey’s correctly identified 99% of the time.Coffee• Discriminate between normal and
decaffeinated.
Examples
Cereals• Flour quality (hardness) although calibration
responded to granular size.• Protein, moisture. Used inline on harvesters.Paper• Determining pulp yield and kappa number for
kraft pulp and black liquor samples.
Future
NIR chemical imaging gives the ability to quantify a chemical component and also provide spatial resolution. – 2D spectroscopy
Future…………..• Fluorescence Spectroscopy• Microbial, bacterial quantification.