Mariachiara Grande

In the Industry 4.0 era, digitalization and automation of large-scale data analysis create new opportunities to enhance productivity, quality, and process efficiency. This paper proposes an integrated framework combining Design of Experiments (DoE), Machine Learning (ML), and Explainable AI (XAI) in a single tool for structured and interpretable data planning, processing, and optimization. A key innovation is a biplot representation that visualizes interactions among process parameters, enabling the identification of optimal variable ranges for ideal process configurations. The framework’s predictions were experimentally validated in laser ablation of glass and polymethylmethacrylate, achieving errors between 0% and 5%. Its modular architecture ensures adaptability across diverse manufacturing processes, offering a generalizable and interpretable approach for intelligent manufacturing. This work demonstrates how combining DoE, ML, and XAI can support data-driven decision-making and advance process understanding within the Industry 4.0 paradigm.

The current level of packaging consumption imposes a need to fabricate single-use food packaging with renewable and compostable materials, such as bio-polyesters (e.g., polylactic acid, PLA and polybutylene succinate, PBS) or cellulose, but their use is still problematic. Fabrication of bio-compostable composites can specifically address impeding challenges, and adhesive lamination, achieved with compostable glue, is becoming more and more popular with respect to the less versatile hot lamination. In this context, plasma activation, a chemical-free oxidation technique of a material’s surface, is used to increase the affinity of three different biomaterials (cellulose, PLA and PBS) toward a compostable polyurethane adhesive to decrease its amount by gluing bio-polyesters to cellulose. Optical Microscopy reveals activation conditions that do not affect the integrity of the materials, while Water Contact Analyses confirm the activation of the surfaces, with contact angles decreased to roughly 50 deg in all cases. Unexpectedly, ζ-potential analyses and subtractive infrared spectroscopy highlight how the activation performed superficially etches cellulose, while for both PLA and PBS, a general decrease in surface potential and an increase in superficial hydroxyl group populations confirm the achievement of the desired oxidation. Thus, we rationalize continuous activation conditions to treat PLA and PBS and to glue them to neat cellulose. While no beneficial effect is observed with activated PLA, bi-laminate composites fabricated with activated PBS fulfill the benchmark for adhesion strength using less than before, while oxygen permeation analyses exclude plasma-induced etching even at a nanoscale.