Streamlined development of continuous blending through integrated process understanding and predictive modeling

Abstract

Within the framework of continuous manufacturing of drug products, continuous blending plays a central role in the quality of the final product. However, the interplay between material properties and process parameters remains complex and system–dependent. This work presents an integrated experimental and predictive framework to characterize and optimize blending performance in a 70–mm GEA linear continuous blender. A full factorial design study was conducted using different raw materials and blend mixtures covering a wide range of powder properties such as flowability, cohesion, and density. The main goal was to quantify how blender speed, throughput, and blender configuration influence hold–up mass (HUM), mean residence time (MRT), number of blade passes (NBP), and time to reach steady state. Distinct blending regimes were identified, with materials exhibiting shear– or transport–dominated behavior depending on their properties and process conditions. A partial least squares (PLS) model was then created to predict HUM from material attributes and process conditions, enabling the estimation of MRT and NBP profiles for new formulations, with only two material characterization tests: bulk/tapped density and shear cell yield locus. Finally, tablet content uniformity studies and residence time distribution studies showed that higher mixing element configurations and adequate hold-up mass are crucial for effective macromixing and reducing throughput variability from the feeders, especially for low-dose formulations

Introduction

Continuous manufacturing (CM) has become a transformative approach in pharmaceutical production, offering better process control, a smaller manufacturing footprint, increased robustness, and greater operational flexibility compared to traditional batch methods. Beyond efficiency improvements, CM is increasingly seen as a way to address systemic issues like drug shortages (Alemie et al., 2025). This technological transition is especially important for tableting, which makes up the majority of oral solid dosage forms (Rogers et al., 2013). In continuous direct compression (CDC), tablet manufacturing generally involves three main unit operations: loss-in-weight feeding of raw materials, continuous linear blending, and compression of the mixture into tablets. Among these, continuous feeding and blending are crucial because they directly affect blend uniformity and, therefore, the content in the final dosage form. Poor blending can cause variability in tablet assay, weight, or mechanical properties, ultimately affecting product quality (Jones-Salkey et al., 2023a).

Mixing performance in pharmaceutical blending is often described in terms of macro-mixing and micro-mixing, both of which are essential for achieving robust and consistent formulations. Macro-mixing controls large-scale material transport and circulation within the blender, helping to reduce upstream feed fluctuations and establish uniform bulk composition. Micro-mixing, on the other hand, occurs at the particle scale and involves local rearrangement, de-agglomeration, and dispersion of individual components. Therefore, while effective macro-mixing ensures overall homogeneity, micro-mixing influences the quality of local composition and is especially important for low-dose formulations and cohesive powders, where small-scale variability can lead to content uniformity failures. Overall blending performance results from a balance between macro-mixing and micro-mixing capacity (Jones-Salkey et al., 2023b).

In continuous blenders, material transport and mixing behavior result from a complex interaction between powder properties, equipment geometry, and operating conditions. A large body of research has shown that blending performance depends on the combined effects of blender speed, line throughput, blade design, and material characteristics. System responses are often measured using metrics such as hold-up mass (or fill level), mean residence time (MRT), residence time distribution (RTD), axial dispersion, and strain-related measures, including the number of blade passes (NBP). Together, these parameters can greatly affect both macro- and micro-mixing performance. Macro-mixing is largely associated with HUM, MRT, and the RTD curve, whereas micro-mixing is more closely linked to agitation intensity and accumulated strain (NBP) (Jones-Salkey et al., 2023b).

Hold-up mass (HUM) is the amount of material inside the blender during steady-state conditions and is directly related to the fill level of the blender. The ratio of HUM to throughput determines the mean residence time (MRT), which indicates the average time the material spends in the blender. Mixing intensity is typically measured using strain-related metrics such as the NBP, which is calculated as the product of blender speed and MRT, representing the total mechanical interactions experienced by the powder. These responses can be calculated as follows.

Another key response in the blending unit operations is the residence time distribution (RTD), which represents the probability of distribution function that describes the amount of time a solid could spend inside the blender. It can be used to predict how fluctuations in a feeder dissipate in a mixer and their subsequent impact on the blend and content uniformity (Tian et al., 2017). Beyond their application to process characterization, RTD models have also been incorporated into predictive and digital-twin frameworks for continuous blending. For example, Beke et al. combined RTD models, feeder characterization, and artificial neural networks to create a digital representation of a continuous blending process capable of predicting process performance and evaluating process robustness under different operating conditions. This work highlighted the potential of integrating process understanding with predictive modeling to support more efficient process development and optimization (Beke et al., 2021).

As explained above, these responses are naturally interconnected: adjustments to a single operating parameter often lead to simultaneous changes in HUM, residence time, and strain, making it difficult to control the mixing duration and intensity independently (Jones-Salkey et al., 2024). Previous studies have shown that tablet content variability (RSD), sometimes considered an indirect measure of micro-mixing, can be minimized by maximizing strain within the blender. This can be achieved by optimizing blender speed and blade configuration to increase strain (Palmer et al., 2020, Van Snick et al., 2017b). However, equivalent residence mass can be achieved through different combinations of blender speed and configuration, indicating that multiple operating pathways can lead to similar macro-mixing conditions but different micro-mixing conditions. For this reason, it is important to take all these factors into account when optimizing the blending process.

Blender speed is a key operating parameter that significantly influences powder blending behavior, as it directly affects particle transport mechanisms and residence time. Multiple studies report that increasing blender speed reduces HUM and, consequently, MRT. Portillo et al. also demonstrated, through positron emission particle tomography (PEPT), that higher blender speeds decrease particle distance travelled inside the blender by limiting the number of loops the particles make (Portillo et al., 2010).

The impact of blender speed, however, is not that straightforward. Several studies have shown that the maximum NBP (strain) can occur at different blender speeds, depending on factors like blender configuration and geometry. For example, Osorio and Muzzio, and Portillo et al. observed that the maximum NBP occurred at higher blender speeds, whereas Van Snick et al. and Vanarase and Muzzio found that, depending on experimental settings, the maximum NBP could be located anywhere within the blender speed range (Osorio and Muzzio, 2016, Portillo et al., 2010; Van Snick et al., 2017; Vanarase and Muzzio, 2011). Van Snick et al. also noted that the blender blade design affects the relationship between strain and blender speed and should therefore be considered a key geometric factor.

Specifically, it has been demonstrated that increasing the number of mixing elements in the blender increases MRT and NBP, leading to improved blend homogeneity for both macro- and micro-mixing. This is because these configurations not only extend the residence time but also expand the shear and mixing zones (Bekaert et al., 2022b; O. Jones-Salkey et al., 2023; Palmer et al., 2020, Portillo et al., 2008; Van Snick et al., 2017; Vanarase and Muzzio, 2011). However, these configurations must be paired with 45° blades to ensure the transportation of material to the blender’s outlet (Portillo et al., 2008, Van Snick et al., 2017a, Vanarase and Muzzio, 2011).

Contrary to blender speed and configuration, throughput has been shown to have a lesser impact on the blending responses. Some studies have reported that the impact of throughput on HUM was marginally significant (Palmer et al., 2020), while others report a negative correlation with residence time, but a positive one with HUM, reflecting higher solids loading and reduced particle–blade interaction time (Jolliffe et al., 2025, Jones-Salkey et al., 2024, Osorio and Muzzio, 2016). However, this inverse relationship between throughput and residence time is primarily observed at low blender speeds, whereas at higher blender speeds, residence time becomes largely insensitive to flow rate as convective transport dominates (Vanarase and Muzzio, 2011).

Similar to process parameters, material properties also influence blending behavior by affecting powder flow, packing, and response to applied shear. As with blender configuration and throughput, the effect of material properties on blending responses, specifically cohesion, decreases as blender speed increases (Jones-Salkey et al., 2024, Portillo et al., 2010). However, at low to intermediate blender speeds, cohesive materials tend to have longer residence times and lower particle mobility (Portillo et al., 2010). Bulk density has also consistently been identified as a key factor determining MRT and fill level, with higher-density materials causing increased HUM and shifting the optimal blender speed for maximum strain toward lower values (Janssen et al., 2023, Van Snick et al., 2017b; Vanarase et al., 2013). More comprehensive studies that included additional material properties also showed that other attributes, such as flowability and compressibility, impact HUM and MRT, thus influencing blend uniformity variability (Bekaert et al., 2022a, Jolliffe et al., 2025).

Taken together, the literature converges on a consistent message: continuous blending performance can be effectively tuned through operating parameters. However, while adjusting these parameters is often the most practical approach, the resulting process understanding is frequently system-specific, limiting the transferability of knowledge across different blender geometries and scales (Jones-Salkey, 2023a).

This paper presents an integrated strategy for optimizing process development by combining process understanding with empirical modeling and direct links to tablet content uniformity. Specifically, the study investigates the blending dynamics of a 70 mm diameter GEA linear continuous blender, examining how material properties and process conditions influence blending performance (HUM, MRT, and NBP). In addition, the time required to reach steady state under various operating conditions is evaluated, an aspect that has not been thoroughly explored in the existing literature. Although this may not be advantageous in a commercial-stage process, it is particularly important at the early stage when material may be scarce, and some trials are still necessary before finalizing the optimal process conditions for a specific formulation.

To further support process development, a partial least squares (PLS) model is developed to predict HUM using only material properties across a range of process conditions, enabling rapid estimation of system behavior without extensive experimental trials. Finally, blending performance is directly connected to downstream tablet quality through content uniformity testing. Residence time distribution parameters and funnel plot analysis are also used to relate process conditions and formulation attributes, including both high- and low-dose blends, to macro-mixing behavior.

Download the full article as PDF here Streamlined development of continuous blending through integrated process understanding and predictive modeling

or continue reading here

Materials

In this study, two raw materials and nine binary blends were used. Since the primary goal was to understand process dynamics across various process parameters and material properties, and to develop a model to predict the HUM for different formulations, we aimed to cover a broad range of material characteristics. For this reason, spray-dried lactose (Supertab 11SD, DFE, Goch, Germany) was chosen as a free-flowing material, while a placebo spray-dried excipient (Placebo 1, Hovione Farmaciencia SA, Lisbon, Portugal) was selected as a cohesive material. Placebo 1 was used as a surrogate for spray-dried APIs because, among the spray-dried dispersions available, it exhibited the highest cohesiveness and poorest flowability, making it a suitable worst-case material for evaluating blender performance. In this study, Placebo 1 and Supertab 11SD will be referred to as M1 and M2, respectively, as described in Table 1. Two binary blends with these materials were also studied: one with 5% w/w and another with 50% w/w of Placebo 1, to simulate and investigate the effect of low-dose and high-dose formulations. These formulations will be referred to as F1 and F2, respectively.

Ana Martins, Cláudia Moura, Ricardo Sousa, João Henriques, João A. Lopes, Streamlined development of continuous blending through integrated process understanding and predictive modeling, International Journal of Pharmaceutics, 2026, 127614, ISSN 0378-5173, https://doi.org/10.1016/j.ijpharm.2026.127614.