The initial sections of Chapter 1 introduce the main goal of this Industrial PhD Thesis and offer an insight on the physico-chemical properties of special clays and their main application fields that are specifically targeted throughout this PhD thesis. The State-of-the-art for machine learning tools in porous materials is also discussed. The latter includes an overview of the datasets, descriptors, and methodology .

Chapter 2 presents this thesis’s achievements, giving a general discussion of the three research articles published within this PhD work. The discussion also discloses a novel practice of machine learning models in capturing the synergy between compounds, which is related to the works of Chapters 4 and 5. The latter is expanded in Annex 1 and is expected to be converted into an additional research publication in the future. A resume of the work realized during the International stay is given and thoroughly presented in Chapter 6.

Chapter 3 includes the “Machine-Learning-Accelerated Multimodal Characterization and Multiobjective Design Optimization of Natural Porous Materials” article published in Chemical Science journal. The contribution demonstrates the first application of machine learning to predict the morphology and surface activity of natural and modified clay-based materials. The suitable proposed material representation allowed us to build predictive algorithms which can be exploited in the design of clay-based acid nanocatalysts.

Chapter 4 moves to “Machine learning-aided design of composite mycotoxin detoxifier material for animal feed”, published in Scientific Reports journal. Random forest was educated by a historical set of in vitro adsorption data to screen and identify the optimal formulation. The latter was directly tested in an in vivo trial in collaboration with the Department of Pathobiology, Pharmacology and Zoological Medicine, Faculty of Veterinary Medicine of Ghent University.

Chapter 5 presents the advantages of coupling the advanced design of experiment tools to machine learning in the context of rheological additive design. The work was presented in the article Datadriven experimental design of rheological clay-polymer composites, published in the Industrial & Engineering Chemistry Research journal. The diversity selection algorithm was arranged to identify formulations correlated to the most different outcomes. Three top promising composites were then prepared and tested in their rheological behavior, validating the proposed approach.

Chapter 6 demonstrated the achievements of the work done in collaboration with the Department of Physics and Chemistry of the University of Palermo during my International stay. The work was completed by running additional experiments in Tolsa laboratories, and the whole story was converted into a manuscript and recently submitted to Microporous and Mesoporous Materials. By using modulated thermogravimetric analysis, the behavior of confined water on the structures of five diverse clays was deeply investigated. The latter allowed us to calculate the activation energy for the removal of the adsorbed water (Ea) and correlate it to adsorption properties. Annex 1 presents an extension of the use of machine learning models in capturing the synergy between components of nanocomposites. The concept was first explored during the development of the work published and reported in Chapter 4.

The predictions of the mycotoxin adsorption followed a non-linear regression within a linear combination of the compounds in the detoxifier formulation. The latter was associated with the ability of machine learning models to consider the synergy between those compounds. The approach was then applied to the work reported in Chapter 5, observing similar behavior. The results will be adapted as a manuscript to be published as a research contribution.