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 [2].
It is clear that, in this case, x1 + x2
+ x3
= 100%. Since
the add-pack treat level is usually prescribed by the manufacturer, we have only one variable to optimise – the VII percentage. Say, we want to get V100 = 15 cSt. Then there is only one solution, x3
= 9.2%.
In a more complex case, where there are 2 variables, we will already get a range of solutions.
This approach can easily be extended to crankcase lubricants, in which case major blend constituents are base oils, additive packages, and viscosity index improvers. 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. An ML-model trained on 1000 blends allows one to predict all essential physicochemical properties with the required accuracy (confidence corridor) from the first attempt in 80% cases, while for an average R&D lab, it will take 5-15 attempts to get there.
Let’s explain how this approach works. Let’s choose the blend viscosity at a given temperature as the property of interest. To be more specific, we take a simplified formulation for SAE 10W-40 motor oil. In the simplest case, it includes only three ingredients: base oil, add-pack and VI improver.
For an N-component blend, we will get an N dimensional graph with two intersecting N-1 dimensional hypersurfaces. Mathematically, we will have to map the N-1 dimensional composition domain onto the M-dimensional property domain. Such a case is more difficult to visualise but computers can handle such tasks with ease. For some properties, there exist rather accurate functional property-composition relationships. Viscosity blending relationships are one good example. For other properties, corresponding functional relationships need to be empirically reconstructed based on available test results. Further, properties such as kinematic viscosity and TBN are required to stay within a certain min-max range, while others – such as Cold-Crank Simulator (CCS) viscosity, Noack volatility, and price – are constrained only from one side.
The AI based formulator assistant platform developed by SBDA comes packaged with an easy-to- use interface customised according to end user demands. No universal interface is feasible at the moment because different blenders use different – and usually confidential – internal processes for purchasing, inventory management, production and product portfolio management, and R&D, which are considered as an essential part of company’s know-how.
Continued on page 22 LUBE MAGAZINE NO.168 APRIL 2022 21
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57
