Learning algorithms for identification of whisky using portable Raman spectroscopy

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Brand substitution and quality control are significant challenges in the high-value spirits industry.
• Accurate and efficient identification methods are crucial for consumer trust and industry integrity.

Purpose of the Study:

• To develop a novel, automated framework for the direct analysis of whisky using spectral data.
• To achieve high accuracy in identifying whisky brands and quantifying key chemical components.
• To demonstrate the practical application of the technique in real-world scenarios, including through-bottle analysis.

Main Methods:

• Integration of a portable Raman spectroscopy device with machine learning models.
• Direct analysis of raw spectral data without human intervention.
• Development of a unique beam geometry for through-bottle spectral measurements.

Main Results:

• Machine learning models achieved over 99% accuracy in identifying whisky brands across twenty-eight commercial samples.
• The framework successfully quantified ethanol concentrations and measured methanol levels.
• Through-bottle spectral measurements demonstrated the technique's real-world applicability and practicality.

Conclusions:

• The developed framework offers a rapid, non-destructive, and highly accurate method for whisky analysis.
• The technique enhances practicality by enabling through-bottle measurements, reducing the need for sample preparation.
• This approach has significant potential for detecting counterfeit or adulterated spirits and other high-value liquids.