Advanced combustion design with coupled unit operations for flowsheet development using DETCHEM andMachine Learning tools

High-fidelity models combined with machine-learning (ML) surrogates offer a powerful framework for accelerating the design of chemical process systems. In this study, independent ML models for two different unit operations (chemical reactor and recuperative heat exchanger) were coupled to optimize a flowsheet. Large datasets were generated using DETCHEM CHANNEL for catalytic methane combustion and a finite-difference model for a microchannel recuperative heat exchanger. High fidelity and fast reactor design tools facilitate data sets of suitable size for ML, with 60,000 model-driven reactor data points established in 18 hours of clock time. Separate ML regression models were trained for each unit operation and then coupled through a thermodynamic fixed-point iteration that enforced consistency between combustor exhaust temperature and heat-exchanger performance. This coupled surrogate enabled rapid system-level evaluation and was embedded within a genetic algorithm to minimize device volume while maintaining thermal effectiveness. The optimized configuration reduced the volume-to-throughput ratio by approximately 60% compared to the initial unoptimized design. Although the heat-exchanger surrogate reproduced its governing model with high accuracy, combustor performance near the light-off transition remained challenging, suggesting that future work should employ physics-informed neural networks to improve robustness and energy-balance consistency.