A bias-corrected ensemble model for quantifying additive content variations in complex lubricant systems.

Area of Science:

• Lubricant analysis
• Spectroscopy
• Chemometrics

Background:

• Lubricant additive concentration is vital for performance and lifespan.
• Detecting trace additives in complex mixtures is challenging due to overlapping spectral peaks and weak signals.
• Accurate quantitative analysis of lubricant additives is crucial for quality control and performance prediction.

Purpose of the Study:

• To develop a rapid and accurate method for detecting trace additive concentrations in lubricating oils.
• To address challenges in quantitative analysis caused by spectral interferences in complex lubricant mixtures.
• To enable precise monitoring of additive levels for optimizing lubricant performance and service life.

Main Methods:

• A physics-informed Bias-Corrected Ensemble Model (BCE) was developed.
• The model utilizes an ensemble of learners based on derivative spectroscopy and the Lambert-Beer law.
• A meta-model compensates for prediction biases from mixing interference, decomposing overlapped spectral features.

Main Results:

• The BCE model accurately detected specific additive content in simulated oil systems.
• The coefficient of determination (R²) for predicting T321 concentration reached 0.949.
• Friction tests on verification oil samples showed errors <5.8% for friction coefficient and <1% for wear scar diameter.

Conclusions:

• The Bias-Corrected Ensemble Model provides accurate quantitative analysis of lubricant additives.
• The method effectively decomposes overlapped spectral features in multi-component mixtures.
• The model's predictions correlate well with tribological performance, validating its practical application.