(11a) Effect of Pyrolysis Conditions on Producing Mesophase Pitch from Varying Ranks of Coal and Correlating Their Product Properties with Machine Learning Models

Coal tar pitch (CTP) is the heavy fraction of the coal tar by-products of coal coking processes and is often used as a feedstock for high-value products, such as high-modulus carbon fiber or needle coke. During the production of these high-value products from CTP, the CTP must undergo conversion to an intermediate product called mesophase. Mesophase formation in carbon materials has been widely studied over the years, and it is well known that mesophase conversion and quality is highly dependent on the chemical properties of the feedstock. Some of the important chemical properties that impact mesophase conversion are aromaticity, heteroatom functional groups, viscosity, and molecular weight. Because these chemical properties critically affect mesophase conversion, some feedstocks are intrinsically more suitable for mesophase conversion than others.

Previous work in our group has shown that feedstocks with chemistry originally not suitable for mesophase conversion (e.g. coal tar from non-coking coals) could be modified to be more suitable for mesophase conversion through the use of controlled secondary gas-phase reactions during pyrolysis. In this work, this approach was investigated more broadly on multiple coals of varying ranks (Wyoming PRB Wyodak, Utah Sufco, Illinois #6, and West Virginia Flying Eagle). The intermediate coal tar products were analyzed for their general chemical functionality through Fourier-transform infrared spectroscopy, and their respective molecular weight distributions were measured through laser desorption ionization mass spectrometry. After converting the coal tar samples into mesophase CTP, the samples were then viewed under a polarized microscope to analyze the mesophase optical textures to determine their mesophase qualities. In addition to investigating this pyrolysis approach on multiple coals, the data collected was used in a machine learning approach to predict mesophase qualities based on the precursor coal tar chemical properties.

Our results show that for all coals investigated, improvements were seen in the mesophase CTP qualities with an increase in the secondary gas-phase reaction temperature and residence time. The most marked mesophase improvements were observed for Utah Sufco, PRB Wyodak and Flying Eagle CTPs, but the Illinois #6 CTP had little improvement. A preliminary model based upon machine learning was constructed using the well-known Elastic-Net regression algorithm from the Scikit-learn Python module. Model training and testing was performed on randomized training/testing data splits at least 50 times and the average training and testing accuracies were 0.819 +/- 0.048 and 0.556 +/- 0.216, respectively.

The results of this work show that the use of controlled secondary gas-phase reactions in pyrolysis may be more broadly applied to various carbon feedstocks to improve their ability to produce desirable mesophase carbon intermediates. In addition, the Elastic-Net model supports that there is a measurable correlation between coal tar feedstock properties and their resulting mesophase CTP qualities. However, the current model displays instability and high variance on the test score, due to the limited dataset collected thus far. Model accuracy and robustness will be improved on the Elastic-Net model once the targeted dataset is complete and smoothing techniques are applied. Other non-linear models, such as random forests and neural networks, will also be explored for comparison on the completed dataset.