Engineering nanocellulose (NC) architectures is fundamentally constrained by the system's profound non-linearity, generating a predictive gap between lignocellulosic deconstruction and the resulting supramolecular assembly. This gap originates from the complex behavior related to lignocellulose raw material, pretreatment, processing, and deconstruction that generates a large surface area and complex colloid chemistry. Deconstruction processes actively encode complex colloidal properties by tuning the non-linear interplay of surface chemistry, anisotropic crystalline facets, and dominant non-DLVO forces. These nanoscale interactions govern all emergent macromolecular behaviors, including self-assembly pathways, gel structure, and viscoelastic rheological properties. Machine learning (ML) has risen as an important tool. Namely, in the field of NC, ML has been exploited to correlate controllable inputs, such as feedstock composition and pretreatment energy, with NC properties in colloidal suspensions and interfaces. By taking advantage of data-driven mapping between process parameters and properties, ML has facilitated the prediction of attributes such as nanofibrillation yield, aspect ratio, viscoelasticity, adsorption and composite strength, complementing physical chemistry insights. There is a prevalence of models that correlate inputs to performance of final NC’s applications, while there is opportunity in developing models for their production, characterization and optimization. In general, decision-tree based models are useful for scarce-data scenarios, whilst more complex algorithms, such as artificial neural networks, can benefit from larger datasets to effectively capture subtle non-linearities. This opinion paper highlights the current trends, opportunities and challenges of integrating artificial intelligence (AI) with the molecular-level understanding of NC interfaces, offering a pathway towards predictive design and upscaling of advanced NC materials.
Signori-Iamin et al. (Fri,) studied this question.