Machine learning to improve multi-dimensional gas chromatography data processing for petroleum samples analysis

Understanding the quantitative chemical composition of petroleum distillate fractions at the molecular level could allow researchers to better control petroleum refining and petrochemical production processes. Multi-dimensional Gas Chromatography (GCxGC) coupled with a wide range of detectors such as flame ionization detectors, sulfur and nitrogen chemiluminescence detectors, and mass spectrometers are powerful tools used to characterize and evaluate hydrocarbon types and hetero-organic compounds of nitrogen and sulfur in petroleum samples. Data processing for these techniques is very time, labor, and expertise intensive due to factors including gas chromatography column deterioration over time causing chromatographic retention time shift, baseline noise, and column bleed.

Automated baseline noise and column bleed filtering and automated chromatographic retention-times alignment were developed using machine learning algorithms to reduce the time spent by analysts to produce finished and fully processed data. These machine learning software features can be easily applied and are applicable to both qualitative and quantitative measurements via multi-dimensional chromatography. The automated chromatographic retention-times alignment and automated background noise filter save roughly 95% and 90%, respectively as compared with manual data processing. These automated functions also improved precision and accuracy in the data processing analysis.

Examples of data processing using both automated background noise filtering and automated chromatographic retention time alignment will be presented.