Industry 4.0 in industrial biorefineries: improving process operations by data-driven and hybrid modeling

Biorefineries are innovative networks of interconnected plants for sustainable production of energy, fuel, and chemicals from renewable biomass. Unlike traditional refineries, they handle diverse raw materials to produce limited products efficiently, minimizing waste. Despite their importance for sustainability, biorefinery implementation faces technical and economic challenges. Research mainly focuses on biomass pre-treatment, upstream, and downstream processing, as well as process synthesis and design. Mathematical modeling plays a crucial role in biorefinery development. Industry 4.0, with advanced data analytics, can improve operations. This Thesis aims to apply process systems engineering and data analytics to support industrial biorefineries, focusing on a pioneering 1,4-butanediol biorefinery in Italy. Two main objectives are pursued: 1. Demonstrate the value of the Industry 4.0 approach for biorefineries. 2. Contribute to the advancement of data-driven modeling. The first objective is accomplished by developing digital systems using advanced data analytics to enhance biorefinery operations. The second objective involves improving existing methods, developing new ones, and proposing guidelines to select suitable models based on data characteristics. The inclusions of domain-specific knowledge in the modelling workflows is of paramount importance for both objectives. To this end, hybrid modeling and feature-oriented modeling are explored. Operations improvement involve analyzing the bioconversion step, using Industry 4.0 to troubleshoot a declining product quality and recover it. In the downstream section, membrane fouling in the ultrafiltration unit is investigated. A soft sensor for membrane resistances is proposed, enhancing fouling monitoring. Feature-oriented modeling identifies process settings related to fouling, aiding maintenance schedule improvement. For advancing data-driven modeling, a method for algebraic inversion of latent-variable models is proposed for product design, retaining all quality variables. A framework for automatic model selection and calibration for fault detection is introduced, requiring no prior assumptions about the nature of faults. These studies illustrate the value of Industry 4.0 and digitalization in industrial biorefineries, providing valuable insights and tools for practitioners. Overall, they are expected to promote the adoption of Industry 4.0 in complex industrial environments like biorefineries.