Developing a surrogate n-hexane pyrolysis reactor model through artificial neural networks learning

A computationally efficient surrogate model for n-hexane pyrolysis was developed using chemical reaction neural network (CRNN), a class of artificial neural networks (ANN) designed to embed chemical kinetics. Conventional chemical simulations require solving thousands of species and reactions, resulting in high computational cost and limited scalability to complex reactor geometries such as naphtha cracking tubes. An idealized plug flow reactor (PFR) simulation of n-hexane and steam thermal decomposition was conducted under ideal gas and constant pressure assumptions to generate time-series training datasets. The training conditions spanned 870–1150 K, 100–300 kPa, and residence times below 0.4 s, mimicking typical naphtha cracking environments. Initial parameter guesses for rate coefficients were derived from detailed kinetic models, and stoichiometric coefficients were constrained to preserve elemental balance. To enhance model generalization, two auxiliary ANN models predicting temperature and residence time profiles were integrated into the CRNN framework, enabling applicability to varying inlet conditions and reactor dimensions. The resulting CRNN-based surrogate model reproduced species concentration profiles with an average relative error below 20.225%, while dramatically reducing computational time per case by over 99.7% and the number of parameters required. This methodology demonstrates the feasibility of CRNNs in modeling complex reaction systems and provides a scalable foundation for process optimization and design in industrial pyrolysis reactors.