Convolutional Neural Networks on Correlation between GC-MS Molecular Data and QCM Gas-Sensing Data

Area of Science:

• Analytical Chemistry
• Materials Science
• Machine Learning

Background:

• Gas chromatography/mass spectrometry (GC-MS) provides detailed chemical composition but is complex.
• Quartz crystal microbalance (QCM) sensors offer real-time gas detection but lack detailed chemical interpretation.
• Bridging these techniques is crucial for advanced gas sensing applications.

Purpose of the Study:

• To establish a correlation between GC-MS compositional data and QCM sensing data.
• To develop a novel methodology for reconstructing GC-MS profiles from QCM sensor signals.
• To link QCM sensor responses to chemically interpretable GC-MS patterns.

Main Methods:

• Development of a one-dimensional convolutional neural network (1D-CNN) model utilizing principal component analysis (PCA).
• Training the 1D-CNN model to predict PCA scores from GC-MS data using QCM sensor signals.
• Utilizing nanostructured QCM sensors (ZnO, SnO2, MgO, TiO2) modified via atomic layer deposition.
• Reconstructing 2D GC-MS maps from QCM time-series data by mapping sensor signals into an invertible PCA latent space.

Main Results:

• The 1D-CNN model achieved high prediction accuracy (average R² = 0.98) for ternary mixtures (ethanol, toluene, dichloromethane).
• Successfully reconstructed full 2D GC-MS maps directly from QCM time-series data.
• Demonstrated a method to link QCM sensor responses to specific chemical peak patterns.

Conclusions:

• A robust methodology is proposed to correlate and integrate QCM sensing data with GC-MS molecular data.
• This approach bridges the gap between different gas-sensing data types, enhancing chemical analysis.
• The findings pave the way for advanced electronic nose systems with improved chemical identification capabilities.