Machine learning and signal processing assisted differential mobility spectrometry (DMS) data analysis for chemical...

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Differential mobility spectrometry (DMS) is crucial for detecting chemical warfare agents, explosives, drugs, and analyzing volatile organic compounds (VOCs).
• Visual analysis of DMS dispersion plots is challenging due to their complexity.
• Existing methods for chemical identification from DMS data require significant experimental data and sophisticated analysis.

Purpose of the Study:

• To develop and validate advanced analytical methods for chemical identification using DMS data.
• To explore the efficacy of machine learning algorithms for differentiating and identifying chemicals in pure and mixed states.
• To investigate the utility of magnitude-squared coherence (msc) as a data-efficient method for chemical composition analysis.

Main Methods:

• Convolutional neural network (CNN) algorithm was employed for chemical differentiation and mixture identification.
• Magnitude-squared coherence (msc) was calculated between DMS data of known and unknown chemical samples.
• Experimental data requirements for both machine learning and msc approaches were compared.

Main Results:

• The CNN algorithm demonstrated high accuracy in distinguishing pure chemicals and identifying components within mixtures.
• Magnitude-squared coherence (msc) proved sufficient for inspecting the chemical composition of unknown samples.
• The msc-based method required significantly less experimental data compared to the machine learning approach.

Conclusions:

• Machine learning, specifically CNNs, provides a powerful tool for accurate chemical identification from DMS data.
• Magnitude-squared coherence (msc) presents a highly efficient and data-minimal alternative for chemical analysis using DMS.
• These advancements enhance the capabilities of DMS for rapid and reliable chemical detection and analysis.