Activated carbon materials have found applications in numerous areas owing to their superior properties. Considering the increasing concerns on climate change and resource depletion,, production of carbon materials from renewable feedstock such as biomass and agricultural wastes have attracted significant research attention. Due to the increasing number of biomass candidates, the development of decision-support tools for biomass screening and process design strategies is highly desirable. Additionally, regardless of the method used to produce carbon material from biomass, the quality of the resultant products commonly depends on several factors including feedstock composition and quality andoperational conditions. HHowever, the effects of different biomass feedstock and process parameters have not been fully understood by previous studies.
To this note, Professor Yuan Yao’s research group at North Carolina State University used a machine learning approach, artificial neural network, to predict the yield and surface area of activated carbon produced from diverse biomass resources and investigate the impacts of biomass feedstock and operational conditions. The first author Mochen Liao is a Ph.D. candidate supervised by Prof. Yao. The second author Prof. Stephen Kelley is the collaborator of this research. The artificial neural network is a data-driven approach for obtaining complex relationships between the inputs and outputs without mathematical description. Despite its applications to in predicting different process parameters of biomass conversion in previous studies, it is yet to be applied in the steam activation process for predicting activated carbon yield and surface area which is the core of the present work. Their work is currently published in the journal, Biofuels, Bioproducts and Biorefining.
The developed multi-layer feedforward artificial neural network models were trained, validated and tested based on the 168 data samples obtained from experimental studies in literature. The extra validation was conducted by comparing the independent experimental data with the results predicted by the trained models. The prediction showed high consistent with the independent experimental data. The model can be used for early-stage screening of biomass feedstock which can be used as a prototype before intensive research. Additionally, the predicted data can be used to support biomass process simulation where a common input needed is the activated carbon yield.
It was necessary to evaluate the impacts of using different combinations of data samples. Three scenarios were developed to investigate the impacts of using different biomass characterization data samples such as proximate and ultimate analysis data. For the activated carbon yield, different data samples did not show large impacts on the accuracy of the trained models. For the surface area, the artificial neural network model using proximate analysis data for biomass composition had the best alignment with the experimental data at the extra validation step. Regarding the prediction of total yield and surface area, the relative importance of the input parameters in terms of feedstock properties, activation conditions, and carbonization conditions were quantified. This can be used to improve the design and operation of the activated carbon production process.
In summary, the North Carolina State University scientists presented artificial neural network models as a powerful tool for the generation of total yield and surface area data to support feedstock selection, process design and simulation for activated carbon production using different biomass feedstock. Therefore, the study will help scientists, engineers and various stakeholders in enhancing experimental and process design for efficient production of activated carbon from biomass.
Liao, M., Kelley, S., & Yao, Y. (2019). Artificial neural network based modeling for the prediction of yield and surface area of activated carbon from biomass. Biofuels, Bioproducts And Biorefining.Go To Biofuels, Bioproducts And Biorefining