Research on characterization of pellet characteristics of straw and potato residue mixture based on deep learning

Compressing straw and potato residues into pellets addresses transportation and storage challenges while promoting sustainable waste management. In this study, a deep learning model was utilized to ascertain the parameters within the microscopic images of the mixed pellets. Moreover, the mechanisms of changes in the molding process of the mixed pellets were analyzed in conjunction with their mechanical properties. Firstly, a cylindrical specimen compressed by an electronic universal testing machine was employed to ascertain mechanical indices, including the relaxation ratio and crushing strength. The ResNet-Unet model was used to identify changes in parameters, such as the presence of solid bridges and cracks, in microscopic images under varying experimental conditions. The findings of the study demonstrate that the model exhibits an accuracy of 92 % and 94 % in identifying solid bridges and cracks, respectively. Secondly, the mechanisms behind the changes in relaxation ratio and crushing strength were subjected to in-depth analysis based on the microscopic images observed by scanning electron microscopy (SEM) under different experimental conditions. The study revealed that the formation of solid bridges during the production of biomass pellets is a dynamic process involving growth, diffusion, and merging with other solid bridges. Additionally, the formation of solid bridges can influence the changes in relaxation ratios and crushing strengths. Meanwhile, the generation of solid bridges was not found to be affected by the feeding amount. It was observed that an appropriate feeding amount could reduce the length of cracks on the pellet surface and improve the quality of the molding.