Artificial neural network for aspect ratio prediction of lignocellulosic micro/nanofibers

In this work a wide sample analysis, under similar conditions, has been carried out and a calibration strategy based on a careful selection of input variables combined with sensitivity analysis has enabled us to build accurate neural network models, with high correlation (R > 0.99), for the prediction of the aspect ratio of micro/nanofiber products. The model is based on cellulose content, applied energy, fiber length and diameter of the pre-treated pulps. The number of samples used to generate the neural network model was relatively low, consisting of just 15 samples coming from pine pulps that had undergone thermomechanical, kraft and bleached kraft treatments to produce a significant range of aspect ratio. However, the ANN model, involving 4 inputs and 4 hidden neurons and calibrated on the basis of pine dataset, was accurate and robust enough to predict the aspect ratio of micro/nanofiber materials obtained from other cellulose sources including very different softwood and hardwood species such as Spruce, Eucalyptus and Aspen (R = 0.84). The neural network model was able to capture the nonlinearities involved in the data providing insight about the profile of the aspect ratio achieved with further homogenization during the fibrillation process.

Automated summary

In this study, a neural network model was built to predict the aspect ratio of micro/nanofiber products. The model, calibrated based on a limited number of samples, demonstrated high accuracy and robustness in predicting the aspect ratio of materials from different cellulose sources.