Pyrolysis Modelling for Process Optimization across Feedstocks and Scales: Principles, Development, Quantitative Validation, and Open Python Tools

Pyrolysis modelling now spans empirical yield correlations, global and distributed kinetics, multistep networks, structural models, detailed chemical mechanisms, particle models and reactor-scale simulations. These approaches resolve different levels of information and are not equally suitable for every feedstock, operating regime or scientific and engineering question. This work critically reviews representative pyrolysis models for lignocellulosic biomass and its major components, manure, polymers including polyolefins, biomass–plastic mixtures, sewage sludge and food waste, with emphasis on their governing principles, development history, application domains, validation evidence and limitations. Selected models and common analysis methods were reconstructed or implemented in Python using source-traceable equations, parameters, units and operating conditions. Experimental and literature evidence is used as the main basis for assessing model applicability. Representative results range from bounded same-study interpolation, such as a biomass product-yield holdout with an RMSE of 2.49 percentage points, to substantially more difficult independent no-refit transfer, such as an HDPE peak-temperature RMSE of 15.74 K. Published reactor-scale validation evidence also shows strong dependence on the predicted output, with food-waste deviations ranging from approximately −7.4% for H₂ and −6.8% for bio-oil to +33% for biochar. Detailed biomass chemistry further demonstrates that correct implementation does not guarantee transferability when feedstock mapping, minerals, interactions or thermal history are insufficiently represented. The combined evidence shows that model fidelity and validation strength are independent. Detailed chemistry is most useful when its additional species, pathways or secondary reactions are required and experimentally supported, whereas simpler empirical, global or multistep models can be more appropriate when they resolve the required engineering quantity within a sufficiently validated domain. Model selection should therefore begin with the required observable, feedstock and operating regime, followed by validation evidence, rather than with maximum model complexity. The associated open Python framework links implemented models with their scientific sources, executable examples, validation records and applicability limits to support reproducible research, model development, engineering analysis and evidence-constrained process optimization.