Driven by the increased consumption of resources, the circular bioeconomy has attained significant attention in recent years. Circular bioeconomy is regarded as a concept of sustainability that limits virgin resource consumption by reducing waste production, reusing and recovering resources, and producing higher value-added materials through waste valorization. For resource recovery, several processes like composting, fermentation, anaerobic digestion, and thermochemical conversion processes are used.

Thermochemical conversion processes are key processes for the valorization of biomass, which are used in the context of circular bioeconomy. Various conventional techniques have been widely used, such as combustion, pyrolysis, gasification, and torrefaction. However, these processes require low moisture content. Compared with those processes, hydrothermal carbonization (HTC) provides a promising potential approach, especially for converting biomass with high moisture content without an initial drying step. Under standard conditions, HTC operates in the presence of water at low temperatures ranging from 180 °C to 250 °C, under autogenic pressures of 2 to 10 MPa, with residence times typically ranging from one to several hours. This study aims to develop a novel methodology that reveals the influence of overlooked process conditions and facilitates the prediction of hydrochar properties during HTC by integrating experimental and computational approaches. As such, the study is structured into two main components. The first component focuses on experimental work to critically assess the commonly used laboratory practices in HTC literature, specifically pre-drying of biomass and to evaluate an often-overlooked process condition, stirring rate, under different process conditions. Building on the experimental findings, the second component focuses on evaluating and developing modelling tools to better understand HTC hydrodynamics and accurately forecast hydrochar characteristics.

The experimental work begins by critically evaluating common assumptions and practices in HTC lab-scale research, with a primary focus on biomass pretreatment. Using three feedstocks, rejected tomatoes, apples, and digestate, the study reveals that pre-drying, even when followed by rehydration, significantly alters HTC outcomes, resulting in 5–10% higher solid and carbon yields compared to naturally wet biomass. This demonstrates that rehydrated pre-dried feedstock is not equivalent to wet feedstock in terms of process products. In addition to biomass pretreatment, the role of stirring was also investigated in this study under both wet and pre-dried conditions, with results showing no significant influence on hydrochar yield or composition. To further explore this, a second set of experiments was conducted using Typha australis, examining the effects of stirring rate alongside temperature, residence time, and biomass-to-water (B/W) ratio. Consistently, stirring was found to have an insignificant impact on key hydrochar properties, such as solid yield, carbon fraction, surface functional groups, morphology, and inorganic behaviour, under the studied typical HTC conditions. These findings challenge the common practices of pre-drying and stirring in lab-scale HTC experiments and call for a re-examination of such practices.

Expanding on the previous experimental insights, the study further investigates the formation of secondary char through a combination of laboratory experiments and numerical modelling. Employing fructose as a model compound, the study examines the effects of stirring across different residence times, revealing that while stirring had no significant influence on pH, mass yield, carbon content, or surface functionality, it did promote the aggregation of microspheres at higher stirring rates, leading to microspheres as large as 70 µm. This behavior was further supported by hydrodynamic modelling, which linked turbulent flow regions to secondary char deposition. To translate the experimental findings into predictive capabilities and support broader process generalization, a data-driven modelling framework was developed. An analysis-ready-database (ARD) was constructed of over 500 data points compiled from original experiments and literature. Decision fusion ensemble machine learning models were trained to estimate hydrochar yield and its elemental and proximate composition. These models achieved adjusted R² values between 0.98 and 1.0 on the test data for predicting hydrochar yield, elemental and proximate analysis.