In lubricating and specialty oil industries, blending is routinely used to convert a finite number of distillation cuts produced by a refinery into a large number of final products matching given specifications regarding viscosity, flash point, pour point or other properties of interest. To find the right component ratio for a blend, empirical or semi-empirical equations linking blend characteristics to those of the individual components are used.
Mathematically, the problem of finding the right blend composition boils down to solving a system of equations, often non-linear ones, linking the desired properties of the blend with the properties and percentage of the blend components.
This approach can easily be extended to crankcase lubricants, in which case major blend constituents are base oils, additive packages, and viscosity index improvers. Artificial intelligence (AI) tools allow accurate predictions of the basic physicochemical properties of such blends. This allows one to speed up formulation development as the number of test blends and the amount of testing can be significantly reduced. Furthermore, formulation price optimization is possible, taking into account available raw material inventories, shared use of certain raw materials across a number of finished products, etc. There also are tools for price capping that allow blenders to counter risks associated with supply disruptions and price volatility.
After completing the “virtual” formulation development, the candidate lubricant properties are fed into the engine tribology simulation block that allows predictions of advanced properties, such as performance in mandatory API and/or ACEA engine test sequences. With the current state-of-the-art, only fuel economy and wear protection can be predicted with sufficient accuracy, for instance the top ring wear (TRW) in Cummins ISM test or Sequence VI FEI. The effect of friction modifiers is factored in using empirical quantifiers for the friction modifier efficacy, depending on which the asperity-asperity friction contribution obtained using the EHD tribology simulations is adjusted. Other difficult to predict properties - cleanliness, cam wear, tappet wear, soot, carbon deposits, etc - require co-processing of large amounts of experimental data and application cases. This is where machine-learning algorithms come handy.
In the present communication, the application of AI tools is demonstrated with a focus on ACEA 2021 engine oil development. The AI Formulator Assistant software developed by SBDA using the industry standard CRISP-DM (CRoss Industry Standard Process for Data Mining) platform keeps record of all tests - including failed ones - and uses this information to continuously improve its predictive power.