Software tools to resolve the unique challenges of molecular mega-models
Date
2019
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
The yield optimization of high value molecular species is a highly pursued objective in the chemical engineering discipline. This optimization has been labeled as “molecular management” and its application to chemical processes is seen as way to increase technology profitability and viability (Speeches, 2018). In theory, molecular management is ideal but in practice it is difficult due the complexity entailed in real-world chemical systems. Actual chemical systems involve large numbers of molecules and or molecular structures impeding the application of molecular management. The structural size of the molecules reacting and the number of reactive sites on the molecules create a combinatorial problem. This problem is further exasperated as increases in molecular numbers are inevitable in chemical reactors. Large numbers of molecules and/or reactions abstract associations with kinetic and process parameters due to sheer numbers involved. Molecular-level modeling can help navigate through these barriers due its fundamental basis and molecular traceability. Molecular-level models can decipher and optimize chemical processes at molecular resolutions. As eluded, molecular-level models of real complex systems can quickly reach large sizes, sometimes reaching mega levels in terms of reactions and/or species. Model development at these scales becomes time consuming and cumbersome affecting all model design phases: building, solving, kinetic parameter estimation, and editing. Advances in computer science allow for the development of new modeling techniques and approaches that decrease the time spent in molecular-level model development of large systems. The Dynamic Model Builder (DMB) is a C++ object oriented modeling framework that accounts for varying model sizes while functioning independently of program compilation. The DMB framework creates, stores, and simulates molecular models from computer system memory. The DMB framework is further enhanced to deal with large systems through parallelization of both the ordinary differential equations (ODE) solver and the kinetic parameter estimation routines. DMB’s implicit ODE solver lower-upper (LU) decomposition routine was parallelized on a CPU-GPU hybrid system using compute unified device architecture (CUDA) based MAGMA GPU libraries. The kinetic parameter estimation objective function simulations were multi-threaded on the CPU using Open Multi-Processing (OpenMP). These parallelization approaches decreased kinetic parameter estimation time for large molecular systems. To illustrate the robustness of this modeling framework three reactor models were investigated: lignin pyrolysis, cellulose pyrolysis and diesel hydrocracking. ☐ A molecular-level kinetic model by means of an adapted Freudenberg lignin structure was developed for lignin pyrolysis at the temperature of 600 °C. A detailed reaction network was established taking into account primary and secondary lignin pyrolysis chemistries from literature. An exhaustive method was produced to handle the large structure reactivity by merging reaction networks. The adapted Freudenberg large lignin structure was modified based on a comparison of the reaction network and experimentally observed products. The evaluation of the kinetic model consisted of validation with molecular species measured in experimental work. The model output showed agreement with experimental results from Zhang et al. (Zhang, Resende, & Moutsoglou, 2014). ☐ Thermogravimetric analysis (TGA) of cellulose pyrolysis was simulated using a molecular-level kinetic model. A temperature ramp between 373-1073 K was imposed in order to achieve the TGA simulation. A recursive optimization method was utilized to calculate the degree of polymerization of the starting cellulose molecule from literature bulk properties. The model’s primary reaction pathways for cellulose decomposition allowed for degradation of the active cellulose molecule by hydrolysis and thermal degradation to create cellobiose, cellobiosan, and glucose. The mechanisms revealed by Agarwal et al. were used to degrade cellobiose (Agarwal, Dauenhauer, Huber, & Auerbach, 2012). An extension of the cellobiose mechanisms were applied to cellobiosan. The reactivity of glucose was captured by the reaction networks proposed in the work of Zhou et al. (Zhou, Nolte, Mayes, Shanks, & Broadbelt, 2014). Monomer and dimer degradation pathways produced volatiles, furans, aldehydes, ketones, and char through a series of complex reactions. Linear free-energy relationships were applied to minimize the number of model kinetic parameters. Executing the cellulose pyrolysis model in TGA mode provided a prediction of the mass loss. The TGA simulation showed agreement with the experimental trends. ☐ A molecular-level kinetic model has been developed for gas phase diesel hydrocracking on a bifunctional metal acid catalyst. The diesel feed was modeled as a complex mixture of paraffins, multi-branched isoparaffins, naphthenics, and aromatics. Linear free energy relationships in the Bell-Evans-Polyani form were applied to minimize the number of model kinetic parameters (Bell, 1936) (Evans & Polanyi, 1935). The model contained 86 independent parameters for control of hydroisomerization and hydrocracking reactions. The model was utilized to develop robust kinetic parameter estimation approaches involving parallel tuning methodologies.