Development of a digital twin framework for biomass pyrolysis: Leveraging machine learning for enhanced fuel...

There is an increasing need for fuel flexibility in the conversion of biomass through thermochemical processes such as pyrolysis. Achieving fuel flexibility only through an experimental approach is both expensive and time consuming. Therefore, a number of computational models are available that can capture the physics of the process. However, most of these models are developed for single-feedstock conditions and they can be quite computationally expensive. An alternative approach to modeling is the utilization of machine learning techniques which can learn the behavior of a process and, once trained using computational or experimental data, they can provide predictions in a timescale of seconds. In this work, a one-dimensional finite volume model was developed to simulate the pyrolysis process in a top-lit updraft (TLUD) biomass pyrolyzer. The model incorporates dynamic porosity change and a global single-step reaction scheme calibrated using experimental results for peach pit pyrolysis. The simulations captured the propagation of a self-sustained pyrolysis front under an opposed-airflow configuration. Data generated with the physics-based model were used to train an XGBoost machine learning model for which a sensitivity analysis was performed to identify the most influential parameters in the pyrolysis process. The trained model was then used to predict pyrolysis front propagation and peak solid temperature in a multi-feedstock experiment using peach and plum pits utilizing a digital twin framework. This work could be utilized to optimize process operating conditions for fuel flexibility, ensuring product homogeneity.