Model-based real-time optimization in continuous pharmaceutical manufacturing

In this work, real-time optimization (RTO) schemes are proposed and applied on a continuous pharmaceutical manufacturing process which consists of three units: synthesis unit, hot melt extrusion unit and direct compaction line. The developed RTO strategies calculate the operating conditions by optimizing the considered objective functions while satisfying the specific constraints. Moreover, the RTO schemes can cope with intentional changes in the process and unintentional changes such as disturbances. Results from simulations and experiments are presented in this work. An advantageous performance is achieved when using the developed schemes.

Highlights

• Real time optimization schemes for continuous pharmaceutical process were developed.
• The developed schemes handle disturbances and the intentional changes of the system.
• The study includes a feeder’s fault case in a direct compaction line.
• The real time optimization strategies improve the considered process behaviour.

Introduction

Continuous manufacturing (CM) has numerous advantages including higher quality assurance, flexibility, faster production and cost reduction (Domokos et al., 2021). This makes CM a promising technology to be applied in the field of pharmaceutical industry to solve the present challenges. For example, this field currently has limitations to increase production promptly in response to emergencies such as pandemics. CM allows to provide more production and capacity by embedding scale-up options (as e.g. increasing flow rates) in the process design and verification (Lee et al., 2015). In spite of that, these privileges of CM technologies are accompanied with essential demands and necessities that emerge every day. One example is the need to optimize the operating conditions to improve the process performance. There are several attempts in the subject of continuous pharmaceutical manufacturing (CPM) that have been conducted to achieve this purpose. For instance, a process for active pharmaceutical ingredient (API) purification was optimized using a flowsheet model in Sen et al. (2013). The system includes crystallization, filtration, drying and API blending with excipient. In Dias et al. (2019), surrogate models and adaptive sampling were used in CPM optimization. Feeders, blender, co-mill and tablet press were employed in this study. In Melkeri (2020), an optimization scheme for feeder refilling was implemented. The method considered the model and a dynamic optimization framework. In addition, an optimization study of a direct compaction (DC) line that consisted of feeders, co-mill, blender and tablet press was presented in Wang et al. (2017a). The refill scheme and the operating condition were optimized in this work.

An optimization study for various API materials in a continuous crystallization process was introduced in Diab and Gerogiorgis (2018). The temperatures and residence times were used as optimization variables. Maximizing API yield in CPM was considered in Diab et al. (2022). Global sensitivity analysis was employed to study the effects of the process parameters on the API yield, which steered the selection of the optimization variables. In Patrascu and Barton (2018), a dynamic optimization study on a continuously-operated pharmaceutical plant was accomplished. The transient behaviour was included in the study and Pareto curves were plotted for productivity and yield. In Chen et al. (2023), energy consumption was optimized for a CPM process. Surrogate models were considered in this work and the results were validated experimentally. Moreover, surrogate-based optimization for continuous tablet manufacturing was studied in Boukouvala and Ierapetritou (2013). The aim of the work was to minimize variability in the properties of the produced material with consideration of the constraints and the product quality. Furthermore, the surrogate-based optimization for CPM was also considered in Wang et al. (2017b) to solve a black-box optimization problem with constraints.

The process conditions were optimized for Ibuprofen continuous synthesis in Patel et al. (2011). A first principles model was used in the optimization. The optimal performance ensured the stability of the process. The selectivity and productivity objective functions for a ketone hydrogenation were investigated in Martinuzzi et al. (2024). The authors calculated the Pareto front of the multi-objective problem. An API synthesis containing four micro-reactors and three micro-separators was optimized in Gerogiorgis and Barton (2009), resulting in improved performance. The continuous synthesis and crystallization of Nevirapine were optimized in Diab et al. (2019). The authors used the interior-point method to solve the nonlinear optimization problem. The published results showed that a higher temperature of the reactor maximized the API yield. In addition, a lower pH was better for the process performance.

Modelling and optimization of an API synthesis reaction network were presented in Grom et al. (2016). The considered operating conditions in the optimization were the temperature, heating rate and the initial concentrations. The study showed that conversions with 91% could be achieved in the second step of the synthesis. However, the trade-off between the yield and productivity was essential. The edaravone API synthesis was optimized with Bayesian self-optimization in Sagmeister et al. (2022). The optimization was multi-objective, with real-time data analysis. The objective functions were maximizing solution and space–time yields. In addition, minimizing equivalents of reagents was considered in this work. The authors used a supervisory control and a data acquisition (SCADA) system (evon XAMControl) with OPC Unified Architecture (UA).

The focus of the previous mentioned studies is to find the optimal operating conditions and not to adapt optimally the inputs of the process to the changes that could happen during the operation such as disturbances. To handle such changes, real time optimization (RTO) is considered as a promising solution. RTO in CPM can be used to modify the set-points during the operation to optimize an objective function while satisfying constraints to handle, for example, changes in product demand or if a fault happened in the system (Nagy et al., 2020). In traditional single-time optimization (non-RTO), the optimal operating conditions of the process are calculated once and do not change during operation. In contrast, RTO recalculates the optimal conditions when changes (such as disturbances) happen in the plant. The RTO frameworks compute the optimal references and provide them to the low-level controllers of the system. Then, the controllers force the outputs of the process to follow the computed references by the actuating signals. Comprehensive reviews that discuss the methods for RTO in general are provided in Chachuat et al. (2009) and Krishnamoorthy and Skogestad (2022). Another suggested solution to deal with faults and disturbances is to use model predictive control (MPC). For instance, MPC was used to manipulate the level of a hopper in a tablet press in Kirchengast et al. (2019). However, the focus of MPC frameworks is mainly to deal with control-based objective functions and not economic ones.

One of the works that uses RTO with CPM was implemented in Singh et al. (2015). RTO was used for a DC line by integrating hybrid MPC-PID with a moving horizon RTO framework. The procedure maximized the profit and minimized the used resources. Another investigation utilizing RTO was conducted in Pineiro et al. (2020). The authors used self-optimizing control in CPM with considering disturbances and noise. The considered disturbances were changes in the molarity of some components and the temperature. The process included mixers, reactors and a separator. Another study that included RTO for CPM was conducted in Kern et al. (2019). The work used nuclear magnetic resonance spectroscopy to provide online data from a reactor for an RTO strategy that used the modifier adaption method (Marchetti et al., 2016). In Hsieh et al. (2024), a feed-forward closed-loop controller was utilized to control a synthesis process of apremilast substance. The controller adjusts flow rates as a reaction to disturbances, so the near optimal yield is sustained. Nonetheless, there are few reports in the literature that consider RTO with CPM and especially with a DC line.

The objective of this work is to develop RTO schemes and apply them on a CPM process that contains a synthesis unit, a hot melt extrusion (HME) unit and a DC line. The main contributions of this paper are:

• A nonlinear model for a blender is integrated in a DC line optimization framework which is able to handle faults in real time. A discharge control system for the waste is included in the strategy. The developed scheme considers the initial level of the hopper in the tablet press by constructing a look-up table to be used online. The work is evaluated experimentally and numerically.
• A cooling and pelletizing system in an HME unit is considered. A model is identified and a PI controller with feed-forward structure is designed to track the diameter and temperature of the strand. The identified model in closed loop with the controller is incorporated within the optimization framework. The optimization scheme considers the change of the strand diameter during the operation and utilizes an economic multi-objective function for the speed of the pelletizer and the air pressure. The study is conducted numerically.
• A chemical synthesis unit is optimized in real time by using a plug-flow reactor model whose parameters are identified experimentally. The optimization strategy considers the change in temperature during the operation and employs a multi-objective function for the product concentration and the inlet flow rate. Also, the scheme incorporated the tuning factor between the considered objective functions. A PI controller with feed-forward structure is designed to track the references from the optimization scheme. The strategy is evaluated experimentally and numerically.

The paper is arranged as follows: Section 2 presents the materials and methods that were used in this work, including the structure of the system, the process details and the control platform, the models and the developed schemes. Section 3 introduces the results and their discussion. The conclusions are provided in the last section.

Download the full article as PDF here Model-based real-time optimization in continuous pharmaceutical manufacturing

or read more here

2.2.2. HME unit and DC line

Salbutamol free-base (Shenzhen Nexconn Pharmatechs Ltd, China) was embedded in the polymeric matrix Eudragit® E PO (Evonik Industries AG, Germany) by means of HME using a Coperion ZSK18 extruder (Coperion GmbH, Germany). Subsequently, the strands were cut into pellets by means of a PRIMO 60 E pelletizer (Maag GmbH, Germany). The produced pellets were composed of 5 wt.% API and 95 wt.% polymer. The compaction of the pellets into tablets was performed using an excipient pre-blend consisting of 82.5 wt.% Tablettose 70 (MEGGLE GmbH & Co. K, Germany), 16.5 wt.% Kollidon VA 64 (BTC Europe GmbH, Germany), and 1 wt.% magnesium stearate (Merck KGaA, Germany).

The continuous DC line consisted of two gravimetric K-Tron KT 20 feeders (Coperion GmbH, Germany). The first feeder was responsible for feeding the pellets, whereas the second feeder was feeding the excipient pre-blend. Subsequently, a Hosokawa Modulomix (Hosokawa Micron Powder Systems, Netherlands) blender mixed these two components. After the blender, a discharge unit was implemented to remove out-of-specification material. Finally, a Fette 102i tablet press (Fette Compacting GmbH, Germany) compressed the blend into tablets.

Hazem Damiri, Martin Steinberger, Lisa Kuchler, Atabak Azimi, Stefano Martinuzzi, Peter Sagmeister, Jason D. Williams, Stefan Koch, Markus Tranninger, Jakob Rehrl, Selma Celikovic, Stephan Sacher, C. Oliver Kappe, Martin Horn, Model-based real-time optimization in continuous pharmaceutical manufacturing, Computers & Chemical Engineering, Volume 192, 2025, 108915, ISSN 0098-1354, https://doi.org/10.1016/j.compchemeng.2024.108915.

See our next webinar:

“Addressing topical drug formulation challenges through excipient selection“

Date: 28th of November, Time: 4:00 PM (Amsterdam, Berlin)

WEBINAR REGISTRATION HERE