Van Krevelen diagrams based on machine learning visualize feedstock-product relationships in thermal conversion processes

Feedstock properties play a crucial role in thermal conversion processes, where understanding the influence of these properties on treatment performance is essential for optimizing both feedstock selection and the overall process. In this study, a series of van Krevelen diagrams were generated to illustrate the impact of H/C and O/C ratios of feedstock on the products obtained from six commonly used thermal conversion techniques: torrefaction, hydrothermal carbonization, hydrothermal liquefaction, hydrothermal gasification, pyrolysis, and gasification. Machine learning methods were employed, utilizing data, methods, and results from corresponding studies in this field. Furthermore, the reliability of the constructed van Krevelen diagrams was analyzed to assess their dependability. The van Krevelen diagrams developed in this work systematically provide visual representations of the relationships between feedstock and products in thermal conversion processes, thereby aiding in optimizing the selection of feedstock and the choice of thermal conversion technique.


Catalytic effect of ash on thermal conversion
Ash in feedstock represents the inorganic residue persisting post-combustion.Comprising elements like calcium, potassium, sodium, magnesium, silicon, phosphorus, sulfur, and chlorine, ash's influence on feedstock's thermal decomposition varies based on its composition and concentration.For example, biomass ash can either catalyze or inhibit specific reactions, potentially acting as a catalyst poison.When a catalyst is introduced to the thermal process, ash can lead to its deactivation through mechanisms such as coking, sintering, and poisoning.Moreover, ash can modify the quality and yield of thermal process products, including bio-oil, gas, and char derived from pyrolysis.Consequently, ash is a pivotal parameter in refining thermal conversion processes and associated catalysts.
In pyrolysis, ash profoundly impacts both product distribution and quality.Ash presence escalates non-condensable gas formation while diminishing yields of organic and aqueous liquid phases.This phenomenon can be traced back to the catalytic actions of ash constituents, notably alkali and alkaline earth metals, which can foster decarboxylation, dehydration, and cracking reactions 5 .
The catalytic influence of ash in gasification accelerates the gasification reaction rate and carboncontaining material conversion by integrating biomass ash or other alkali metal compounds as catalysts.
Rich in alkali metals like potassium and sodium, biomass ash can reduce activation energy and enhance the gasification process's reactivity.The catalytic impact of biomass ash is contingent upon various factors, including the biomass ash addition ratio, alkali metals' chemical forms in the ash, the fuel's mineral content, and the gasification temperature and environment.The underlying catalytic mechanism involves the creation of a molten or semi-molten alkali metal compound layer on the fuel char's surface, streamlining the reactant and product transfer between gas and solid phases.This catalytic influence can augment gasification's efficiency and selectivity, potentially decreasing the operational temperature and capital expenditure 6 .While ash's catalytic effect in other thermal conversion processes isn't exhaustively discussed here due to length constraints, its significance in thermal conversion is undeniable.However, this study predominantly centers on the H/C and O/C ratios of the feedstock.Delving into ash's effects will be a subsequent focus in our machine learning research.Supplementary Figure 2 presents the range of the ash content in each database.
Supplementary Figure 2. The ash content in the feedstock in the database for each model.feedstock's contents of lipids, proteins, polysaccharides, lignin, cellulose and hemicellulose.For instance, biomass with high cellulose, hemicellulose, and polysaccharide composition might produce furfurals (throughout dehydration route) at HTC process temperatures higher that 200°C 13 .Furfurals are considered one of the major platform chemicals that are used in the polymer industry to produce plastics and adhesives, as well as for the production of inks, fertilizers and flavoring compounds.In addition, can be used as a solvent to produce tetrahydrofuran (THF) and levulinic acid (LA) 14 .Sewage sludge due its particular mixing contents of polysaccharides, lipids, and proteins, can promote the formation of organic acids such as amino acids, volatile fatty acids, and nitrogenous compounds such as ammonia 15 .Volatile fatty acids are valuable in the food industry and ammonia can be used as a hydrogen carrier or either as a precursor for the production of fertilizers.
for the pyrolysis bio-oil models of H/C (0.599) and HHV (0.666).Hence, most of the created models can predict the output variable to a certain degree of accuracy.
Another critical parameter we consider is the weights of the importance of the feedstock's H/C and O/C in relation to the total importance of all input variables.The higher the importance of the feedstock's H/C and O/C, the greater impact these variables have on the output, and therefore, the higher reliability of the diagram.The sums of the importance percentages of feedstock's H/C and O/C in each model are also presented in Supplementary Figure 6.Some models show a very low weight for the sums of H/C and O/C, such as all of the pyrolysis char models with a sum lower than 10%.Although the input importance can only reflect the influence of an input variable on the output in a specific ML model, it can still indicate the importance of the input parameter to some extent in the real situation.For example, the importance percentage of the input variable feedstock's N in the pyrolysis char-N model is 76.8% (Supplementary Figure 15), while the sum of feedstock's O/C and H/C is only 8.5%.It is reasonable to assume that the N content of the char is mainly determined by the initial N content of the feedstocks, while the O/C and H/C can only influence the N content by affecting the relative H and O contents.Based on these, we believe the very low sums of the importance percentages of O/C and H/C inputs in the pyrolysis char models contribute to the unusual patterns observed in the pyrolysis char diagrams.We also do not recommend relying on some of the other diagrams that show reasonable trends but have a low value for this sum.