Enhancing Torrefaction Process Modeling with Data Augmentation: a Study on Agri-Food Residues

Agri-Food residues can pose environmental challenges, but through proper valorization, they can support circular economy principles by providing alternatives to fossil fuels. This study focuses on modeling the torrefaction process of an industrial hazelnut processing waste (i.e., roasted cuticles) to produce high-density solid biofuels. An Artificial Neural Network (ANN) model is proposed. The work leverages the innovative approach of data augmentation through Gaussian noise addition to a relatively scarce experimental dataset. The expanded dataset greatly improved the Neural Network prediction performance: the method achieved considerably lower mean square error compared to the model output based on experimental data only. This approach offers several benefits: it enables accurate modeling with limited experimental data, reduces the need for extensive experimental work, lowers costs, and improves process optimization capabilities, presently in torrefaction processing and, more in general, in resource utilization and waste management.