Abstract
Rationale
In the lubrication industry, commercial base oils are commonly made up of blends of base oil stocks from different sources in different ratios to reduce production costs and modulate rheological properties. This practice introduces complexity in lubricant design because as the chemistry of the base oil becomes more complicated, it can become harder to formulate the base oil – particularly when the ratio of the original base oil stocks is unknown.
Methods
In this study, field ionisation mass spectrometry is used to collect chemical information on a range of base oil mixtures. The resultant data are processed within the Python workspace where molecular formulae are assigned to the components and statistical analyses are performed. A variety of regression techniques including regularised linear models and automated machine learning are evaluated on the data.
Results
The use of an automated machine learning pipeline yields insight into effective modelling strategies that could be applied to the data obtained. The best results were obtained using polynomial feature generation combined with ridge cross-validation regression. Overall, with this methodology it is possible to resolve the ratio of group 2 and group 3 base oil within a blended mixture to an accuracy of ±5%.
Conclusions
The strategies outlined in this study show how modern data science and chemometrics can be applied successfully to resolve the ratio of a complex mixture.
In the lubrication industry, commercial base oils are commonly made up of blends of base oil stocks from different sources in different ratios to reduce production costs and modulate rheological properties. This practice introduces complexity in lubricant design because as the chemistry of the base oil becomes more complicated, it can become harder to formulate the base oil – particularly when the ratio of the original base oil stocks is unknown.
Methods
In this study, field ionisation mass spectrometry is used to collect chemical information on a range of base oil mixtures. The resultant data are processed within the Python workspace where molecular formulae are assigned to the components and statistical analyses are performed. A variety of regression techniques including regularised linear models and automated machine learning are evaluated on the data.
Results
The use of an automated machine learning pipeline yields insight into effective modelling strategies that could be applied to the data obtained. The best results were obtained using polynomial feature generation combined with ridge cross-validation regression. Overall, with this methodology it is possible to resolve the ratio of group 2 and group 3 base oil within a blended mixture to an accuracy of ±5%.
Conclusions
The strategies outlined in this study show how modern data science and chemometrics can be applied successfully to resolve the ratio of a complex mixture.
| Original language | English |
|---|---|
| Article number | e9214 |
| Number of pages | 8 |
| Journal | Rapid Communications in Mass Spectrometry |
| Volume | 36 |
| Issue number | 1 |
| Early online date | 8 Nov 2021 |
| DOIs | |
| Publication status | Published - 15 Jan 2022 |
Bibliographical note
Funding Information:This work was funded jointly by the EPSRC (Swindon, UK) and BP Technology Centre (Pangbourne, UK) as an iCase scholarship awarded to SE (voucher number 1522‐0050).
Publisher Copyright:
© 2021 The Authors. Rapid Communications in Mass Spectrometry published by John Wiley & Sons Ltd.
Keywords
- base oil mixtures
- regression
- chemometrics
- automated machine learning