Pittcon 2022 | March 5 - 9, 2022
Georgia World Congress Center
Atlanta, Georgia, USA

ALL TIMES SCHEDULED ARE EASTERN STANDARD TIME (EST)

Preprocessing, How To Do and Not To Do

  • SC Office A302
  • Course Number:SC14484
Sunday,March 06, 2022:8:30 AM -12:30 PM

Speaker(s)

Instructor
Franklin Barton
Dr.

Description

(Laptop Required)  The first things one learns when applying NIR spectroscopy and chemometrics to analytical problems is that you need to “preprocess the data”. If spectra were perfect and spectrometers produced identical spectra of identical samples and all reference data represented the analyte’s concentration in the spectra, then preprocessing would not be needed. Baseline offsets, particle size, heterogeneity and the vagaries of analytical procedures on different samples means it’s necessary to remove any interferences that do not belong to the analyte’s spectrum. When preprocessing is done correctly the variability is removed from the spectra but when done incorrectly, can introduce features into the spectrum that could be interpreted as actual chemical signatures. This course will look at the common preprocessing techniques and how they impact the spectra and the models. We will be providing spectra and the ability to practice these techniques so each student should bring a laptop.

Register for this Short Course


Track(s)


Course Outline

1. What are we measuring? The amount /concentration of some analyte
2. Relating the spectra and reference value by some algorithm
3. What is a perfect spectra?
a. High S/N
b. baseline separation of bands
c. known absorbance and extinction coefficient
d. “Accurate” reference data (precision vs accuracy)
e. Solution sample transmission mode
f. Construct Beer’s Law plot
4. Things aren’t perfect
a. A host of sampling issues
b. Particle size
c. Color (index of refraction)
d. Grinding vs. As Is
e. Where to sample
f. When to sample
g. Why sample
5. Preprocessing removes things not associated with the spectrum (supposedly)
a. Early case baseline correction
b. Multiplicative effects
c. Offsets
d. Varying particle size
6. What algorithms take care of which features?
a. Derivatives
b. SNV (Autoscale, Normalize)
c. MSC
d. What works
7. Transformations, Filtering, Normalization and Scaling and Centering

Learning Objectives

To appropriately apply mathematical preprocessing to spectral data
Enhance the Predictive capability of Chemometric models
Build Robustness into models

Additional Info

Categories (Up to 5):
Analytical Metrology,Chemometrics,Spectroscopy

Proposal Number:
SC14484

Scientific Specialties:
Spectroscopy - IR/Raman,Spectroscopy UV/Vis

Target Audience:
This course is designed for anyone new to near infrared and uses with chemometric models. It is also designed as a refresher for those who have been away from the field for a while. The course is designed for the user to understand what is going on rather than the mathematical aspects of the pre-processing techniques.

Type of Industry:
Agriculture,Analytical Chemistry/Equipment/Instruments,Biotechnology/Biochem,Contract Laboratory,Cosmetics/Fragrances/Flavors,Education,Environmental/Air/Water/Wastewater,Food/Beverage/Consumer Products,Instrument Manufacturing,Petroleum/Petrochemicals,Pharmaceutical,Polymers/Coatings/Rubber