Food Analysis Lab · W10

Vibrational spectroscopy basics & spectrometer calibration

Vibrational Spectroscopy (NIR vs Raman) & Spectrometer Calibration

Food Analysis Lab · 4-hour session
Five NIR models, one fixed per group · two Raman units (532/785 nm) shared in rotation · calibration · pure-substance spectra → class spectral library

NIR overtone/combinationRaman shiftSilicon wafer/polystyrene calibrationWhite reference/dark currentFluorescence background

Before we start, watch a factory video

Factory NIR “measures” protein in a few seconds?

Flour mill NIR line (15:11–16:14, reports protein/moisture/ash within seconds) ↗

Think first: when an NIR instrument screen shows “Protein 12.3%”, did it measure protein molecules directly? Or did it measure something else and then convert? If it converts, what is the conversion based on? This lesson answers that question — and it is also why the first half of this course teaches “reference methods” such as Kjeldahl.

Today’s task: understand how NIR and Raman vibrational spectroscopy differ, calibrate your group’s instruments, then measure a set of pure-substance spectra. These spectra become the class-wide spectral library and the reference data for the W15–16 sweetener project.

Today’s learning goals

After W10 you should be able to…

Explain that the NIR signal comes from overtones and combinations (overtone/combination) of X–H bonds, and why the bands are broad and overlapping, so chemometrics is needed
Explain that the Raman signal comes from inelastic scattering, define Raman shift (cm⁻¹), and say why spectra are sharp like fingerprints
Explain why water is a strong NIR absorber but a very weak Raman scatterer
Explain why colored/fluorescent samples are often measured at 785 nm, while 532 nm gives stronger scattering but also a larger fluorescence background
Convert wavelength (nm) ↔ wavenumber (cm⁻¹), and state how transmission absorbance A and diffuse reflectance R are computed from dark current and white reference

Complete the Raman x-axis calibration (silicon wafer, polystyrene) and record the results for each of the two lasers
Complete the NIR dark current / white reference / wavelength check, and recognize the baseline anomaly caused by an InnoSpectra cartridge installed the wrong way round
Complete 5 replicate scans, read the mean spectrum and SD spectrum, and find where the noise is largest
Follow laser safety and solvent-handling rules, measure your group’s pure-substance spectra, and upload them using the naming rule

Theory review: this lesson does not repeat the full theory of infrared spectroscopy and spectroscopy in general — see 📖 Full theory deck (Nielsen Ch8, infrared spectroscopy) → and 📖 Full theory deck (Nielsen Ch6, spectroscopy overview) →.

Spectroscopy crash course

Wavelength (nm), wavenumber (cm⁻¹), energy: three quantities that convert into each other

ν̃ (cm⁻¹) = 10⁷ / λ(nm) E = hc/λ

Worked examples: λ=1450 nm → ν̃ ≈ 6897 cm⁻¹; λ=785 nm → ν̃ ≈ 12739 cm⁻¹ (the same light described on a different scale; wavelength and wavenumber run in opposite directions: longer wavelength, smaller wavenumber)
Today’s range: near-infrared (NIR) 780–2500 nm ≈ 12800–4000 cm⁻¹; mid-infrared (MIR) 2.5–25 µm (2500–25000 nm) = 4000–400 cm⁻¹. Today’s NIR and the Raman later all live in this conversion system
E=hc/λ: h is Planck’s constant, c is the speed of light, λ is the wavelength; the shorter the wavelength (the larger the wavenumber), the higher the photon energy. We will use this when we discuss Raman fluorescence below (a 785 nm photon has lower energy than a 532 nm photon)

Principle

Molecular vibration = a spring: why overtones and combinations exist

Think of a chemical bond as a spring: the atoms vibrate back and forth along the bond, but quantum mechanics allows only specific vibrational energy levels (v=0,1,2,3…)
The fundamental (v=0→1) signal is the strongest and falls in the mid-infrared (MIR). Real molecules are not perfect springs (anharmonic), so v=0 can also jump directly to v=2 or v=3 with small probability. These are the first overtone (≈2ν), the second overtone (≈3ν), and the combination (ν₁+ν₂) of two vibrational modes added together, which happen to fall in the NIR
Each step up makes the signal about 10–100× weaker. That is why NIR absorption is weak, light can penetrate a few millimeters into the sample, and no dilution is needed and the method is nearly non-destructive. X–H bonds (hydrogen is the lightest atom) have the highest vibrational frequencies and the strongest anharmonicity, so their overtones and combinations fall right in the NIR range

What NIR measures

Two measurement modes: transmission vs diffuse reflectance

💧

Transmission (liquid samples)

Light passes straight through the sample; we measure incident light I₀ and transmitted light I:
A = log₁₀(I₀/I) = εbc (the same Beer-Lambert law as the absorbance equation you learned in W07 ↗, just at NIR wavelengths; ε = molar absorptivity, b = path length, c = concentration)

🧂

Diffuse reflectance (powders / opaque samples)

Light hitting a powder surface is scattered repeatedly, and the detector signal is converted to reflectance:
R = (sample − dark) / (white − dark). A powder has no single path length, so apparent absorbance A = log₁₀(1/R) is used in place of true absorbance for chemometric modeling

How to measure “dark” and “white”, and how often, is practiced in “Calibration 2” below.

Principles (brief)

NIR: overtones/combinations vs Raman: inelastic scattering

NIR: the signal comes from overtones/combinations of X–H bonds such as O–H, C–H, N–H, e.g. O–H (~1450, ~1940 nm) and C–H (~1150–1210, ~1700–1760 nm). Peaks are broad and overlapping, so quantification always needs chemometrics; water is a very strong absorber and often dominates the spectrum; non-destructive, done in seconds, widely used for real-time production-line testing
Raman: the inelastic scattering of laser light by the sample. The wavenumber difference between outgoing and incoming light is the Raman shift (cm⁻¹). The spectrum is sharp like a fingerprint and can be matched directly against a database to identify a substance
Water is a very weak Raman scatterer and barely interferes, a big advantage for aqueous samples. The biggest enemy is the fluorescence background (details on the next slide)
The virtual state in the diagram is a “non-real” energy level that appears while the laser photon briefly interacts with the molecule. The molecule does not actually stay there; it is just how quantum mechanics describes that instant of interaction

Principle

Two Raman problems: weak signal and fluorescence swamping

🎲

Raman is naturally rare

Only about 1 in every 10⁶–10⁸ incident photons undergoes inelastic (Raman) scattering; nearly all the rest are elastically scattered (Rayleigh) or pass straight through, so the Raman signal is inherently very weak

📉

Intensity ∝ 1/λ⁴

Scattering intensity is inversely proportional to the fourth power of the laser wavelength: (785/532)⁴ ≈ 4.7, so under the same conditions a 532 nm laser gives a Raman signal about 4.7 times that of 785 nm

🔥

Fluorescence = real absorption, then re-emission

Electrons in the sample truly absorb photons and jump to an excited state, then emit broadband fluorescence that can be thousands of times stronger than the Raman signal and completely bury the Raman peaks. It is a completely different phenomenon from Raman scattering

🔻

785 nm photons have lower energy

E=hc/λ: the longer the wavelength, the lower the photon energy, so it is less likely to push electrons to an excited state and therefore less likely to cause fluorescence. This is why dark or fluorescent samples are often switched to 785 nm, even though the signal itself is weaker

Mini-game ①Is this property NIR or Raman?0 / 8

Sort the 8 statements below into “NIR properties” or “Raman properties”.

Data · simulated spectra

Simulated Raman spectra: 532 nm vs 785 nm, same peak positions

Illustrative data / simulated spectra (not measured by this class); the x-axis is evenly spaced Raman shift (cm⁻¹). The 532 nm scattering signal is stronger, but the fluorescence background is also much higher and buries the small peaks; the 785 nm background is flat and the peaks are clear. This is why colored/fluorescent samples are often switched to 785 nm. Dark samples containing conjugated pigments (caramel color, carotenoids, chlorophyll) fluoresce most easily; white crystals and colorless clear liquids interfere little.

Data · simulated spectra

Simulated NIR absorption spectra: water vs ethanol

Illustrative data / simulated spectra, not the specification of any particular instrument. The two water peaks fall near 1450 nm (O–H first overtone) and 1940 nm (O–H combination). Besides a weaker O–H peak, ethanol also shows a C–H first overtone at ~1700–1760 nm and a C–H second overtone at ~1150–1210 nm. This is why the absorbance at a single wavelength can hardly tell the two apart; you need the shape of the whole spectrum (chemometrics).

Bridge

What is chemometrics? — W11–W16 roadmap

Chemometrics = using statistical methods to extract information about chemical composition/properties from a large body of (multivariate) spectral numbers
A spectrum is just a pile of numbers (intensity vs wavelength/wavenumber); the instrument does not know “this is protein”. You must first measure the true value with a reference method (e.g. Kjeldahl for protein W5 Kjeldahl ↗, titration for acidity W4 Titration ↗), so that a statistical model can learn the relationship between “spectral shape” and “that value”. This is called calibration
The next six weeks: preprocessing cleans the spectra → PCA to see what the data look like → PLS to build a quantitative model → VIP / classification / outlier detection → design of experiments (DOE) to plan samples → the mixture-resolution challenge

Calibration 1

Raman x-axis calibration: silicon wafer · polystyrene · cyclohexane

💎

Silicon wafer

A single sharp peak at ~520.7 cm⁻¹; check whether the peak position falls within the instrument’s tolerance — the most common reference for x-axis calibration

🧴

Polystyrene

Main calibration peak 1001.4 cm⁻¹ (ring-breathing mode); 620.9, 1031.8 and 1602.3 cm⁻¹ allow multi-point cross-checks

🧪

Cyclohexane

Peaks such as 801.3 cm⁻¹, used according to the ASTM E1840 Raman shift standards guide

ASTM E1840 = the Raman shift standards guide; it lists accepted peak positions for silicon wafer / polystyrene / cyclohexane etc., so different instruments align to the same wavenumber scale. 532 nm and 785 nm are two separate lasers, so each needs its own x-axis calibration; record laser power and focus (too much power can burn/degrade the sample; poor focus weakens the signal).

Calibration 2a

NIR: dark current · white reference · wavelength check

Warm-up: follow the instrument’s recommendation and let the light source and detector stabilize before measuring
Dark current: the detector background signal with the light path fully dark; the first step of every calibration
White reference: e.g. Spectralon/PTFE (~99% reflectance); redo it every time before using the instrument / before each batch of samples to set the reflectance baseline
Wavelength check: use polystyrene or a NIST rare-earth oxide reflectance standard (e.g. the SRM 1920 series, if available) to confirm wavelength accuracy
Measure a known spectrum: e.g. water, to confirm the baseline and band shape are normal before measuring unknown samples (a real cautionary example on the next slide)

R = (sample − dark) / (white − dark) A = log₁₀(1/R)

Calibration 2b

A real QC lesson: InnoSpectra cartridge installed the wrong way round

The real lesson: if the 1 mm cuvette cartridge is installed in the normal orientation, the cartridge base blocks the lower light source, so the baseline rises and the expected features almost disappear; only installing it the other way (inverted) gives the correct spectrum. That is exactly why the QC habit is to always measure a known spectrum (e.g. water) first and confirm a normal baseline before trusting any data that follows.

Repeatability

What to look at in replicate scans: SD and SNR

SD (standard deviation) — already covered in W01 ↗; it describes how spread out the values are when the same sample is measured repeatedly
SNR (signal-to-noise ratio) = signal size ÷ SD of the noise; the larger the SNR, the cleaner the signal relative to the noise
The two ends of a spectrum usually have lower light throughput (poorer source/detector efficiency) → weaker signal → usually lower SNR; but whether the absolute SD also rises depends on the noise source and the scale shown, and cannot be inferred directly from “weaker signal”. You will see this in the replicate-scan plot on the next slide
Averaging multiple scans reduces noise: averaging n scans reduces random noise to about 1/√n of the original; averaging 5 scans → noise falls to 1/√5 ≈ 1/2.2, an improvement of about 2.2×

Data · repeatability

Replicate scans: 5 scans → mean spectrum and SD spectrum

Illustrative data / simulated spectra; the x-axis is evenly spaced Raman shift (cm⁻¹). The SD (shown as the two “mean ± 1 SD” lines) is largest at the two ends of the spectrum: the edges usually have lower light throughput and poorer signal-to-noise, so repeatability is least stable there; the middle region with stronger signal is steadier.

Safety rules

Laser safety · solvent handling

🚫

Class 3B/4 laser: never look into the beam

Use goggles or the instrument enclosure as required; with the laser on, do not put your eyes near the beam path

💍

Avoid reflective jewelry when operating the laser

Rings, watches, metal bracelets and other reflective items can cause stray reflections; remove them before you start

🧴

Handle ethanol / acetic acid in a ventilated area

Avoid breathing the vapor or skin contact; cap the cuvettes after use and put them back

🧹

Clean up powder / liquid spills immediately

Handle spills per the lab SOP to avoid cross-contaminating the next group’s measurements

Mini-game ②Order the instrument start-up and calibration steps— / 6

Use ▲▼ to put the 6 steps for “before starting the day’s measurements” in the correct order.

Mini-game ③Is this sample likely to cause fluorescence interference?0 / 6

Sort the 6 samples into “likely to cause fluorescence interference” or “unlikely to cause fluorescence interference”.

Mini-game ④NIR / Raman principles and calibration: live quiz0 / 5

Lab work

Pure-substance spectral measurement → build the class spectral library

Powder samples (cuvette / powder holder): sucrose, aspartame, sucralose, erythritol, maltodextrin
Liquid samples (cuvette tube): water, ethanol, acetic acid
Measure every sample with your group’s fixed NIR + the 532 nm and 785 nm Raman units (in rotation)
These spectra become the W11 class spectral library and the reference spectra for the W15–16 sweetener project

532 vs 785 comparison experiment: pick a fluorescent sample (e.g. brown/black sugar, or unrefined cooking oil), measure it at both 532 nm and 785 nm, and overlay the spectra to compare the fluorescence background. This is a hands-on check of today’s “785 nm reduces fluorescence” principle on a real sample.

Mini-game ⑤Decision challenge: calibration and QC interpretation0 / 5

Mini-game ⑥Calculation challenge: Raman shift → absolute scattered wavelengthUnanswered

785 nm laser excitation, silicon wafer 520.7 cm⁻¹ peak. Find the absolute wavelength (nm) of the scattered light. Formula: 1/λ_scattered = 1/λ_laser − Δν̃ (convert Δν̃ to cm⁻¹, convert wavelength to cm and take the reciprocal).

4-hour lab schedule

Principles + safety → calibration → measurement → repeatability → upload

Time	Contents
0:00–0:50	Lecture (wavelength/wavenumber conversion, NIR vs Raman, chemometrics roadmap) + laser and solvent safety demo
0:50–1:30	Raman wavenumber calibration (silicon wafer, polystyrene; once each for 532/785) and NIR dark current / white reference / wavelength check
1:30–3:20	Pure-substance measurement: two Raman units in rotation for the whole class; each group measures with its own fixed NIR at the same time
3:20–3:45	Pick one sample and do 5 replicate scans; plot the mean spectrum and SD spectrum
3:45–4:00	File naming and upload (naming rule: group_instrument_sample_replicate.csv)

Homework

File upload and calibration record

📁

Upload spectra following the naming rule

group_instrument_sample_replicate.csv — organize and upload all of today’s pure-substance spectra (including the 5 replicate scans)

📝

One-page calibration record sheet

One page per instrument: standard, expected value, measured value, pass/fail (Raman x-axis, NIR dark current/white reference/wavelength)

🗂

Check spectral library completeness

Confirm every sample and every replicate has been uploaded; this is the foundation for building the class spectral library in W11

🔬

Prepare for W15–16

These pure-substance spectra are also the reference spectra for the sweetener project; organize the files and label them clearly now

Today’s uploaded spectral library will be used in W11 for preprocessing and similarity matching in Colab NB09 ↗.

Self-check

By the end of today you should be able to…

✓State that the NIR signal comes from overtones/combinations of X–H bonds and needs chemometrics for quantification
✓State that the Raman signal comes from inelastic scattering, and Raman shift (cm⁻¹) is independent of laser wavelength
✓Explain why water is a strong absorber in NIR but a weak scatterer in Raman
✓Explain why colored/fluorescent samples often use 785 nm rather than 532 nm
✓Convert wavelength ↔ wavenumber, and state how A and R are computed from dark current/white reference
✓Complete the Raman x-axis calibration (silicon wafer, polystyrene) and record the results
✓Complete the NIR dark current/white reference/wavelength check and recognize the cartridge-orientation anomaly
✓Complete 5 replicate scans and read the mean spectrum and SD spectrum
✓Follow laser and solvent safety rules and upload spectra according to the naming rule

Tap to tick — make sure you really can do it.