(442a) Simulating the Glass Transition of Steam Cracked Tar

Molecular modeling tools, like Molecular Dynamics, provide a unique perspective on microscopic properties and processes. While there has been a lot of effort using Molecular Dynamics to simulate oil, there has been a reoccurring concern: “Do the molecules being simulated accurately represent the oil’s composition?” While there has been significant progress towards determining the structure of various components in crude, there are still challenges in determining the whole composition accurately. In addition, there are practical concerns given that Molecular Dynamics can only simulate a relatively small sampling of the great diversity of molecules that exist in a typical crude. Significant progress has been made to mitigate these problems, with a great example being the development of “Digital Oil”[1]. Using a variety of analytical techniques, representative molecules are generated to fit available experimental data. The number of molecules used is chosen by the modeler based on how many molecules they wish to simulate and how many are required to sufficiently fit the data.

This route for producing Digital Oil is very promising. However, this method—like any Molecular Dynamics based approach—is still limited by the fact that a small subset of molecules must be chosen from the whole crude. The goal of the work being presented here is to provide a proof of concept for an approach that captures the sensitivity of the predicted property to the particular molecules being selected. This is useful when there is uncertainty regarding how many structures are actually required to precisely calculate a property of interest.

In this study, we investigated a model system involving different solubility fractions of steam cracked tar with different glass transition temperatures. Using a variety of analytical and compositional modeling techniques, a composition of well over 2,500 different molecules was generated for each sample. In an automated workflow, force field parameters were generated for each of these molecules, and then molecules were randomly selected to generate multiple simulations for each sample allowing confidence intervals to be determined. Calculated glass transition results were then compared to experimental data generated by differential scanning calorimetry. There was a good match between calculated and experimental results for most samples, with the few outliers likely being caused by flux contamination that was not included in the composition model. These positive results indicate that this method is a promising tool to virtually probe the properties of crude-based materials. This may be especially helpful for developing novel processes where limited experimental data is available, such as creating higher value carbon-based products from petroleum rather than fuel.

References

[1] M. Iwase; S. Sugiyama; Y. Liang; Y. Masuda; M. Morimoto; T. Matsuoka; E. S. Boek; R. Ueda; K. Nakagawa, Energy Fuels, 32 (2018) 2781.