Advanced Modeling, Simulation and Optimization for In Silico Process Design

Research output: Book/ReportPh.D. thesisResearch

48 Downloads (Pure)


Today, the pharmaceutical industry faces a various number of challenges including high competition with generic market upon patent expiration, pricing pressures, higher costs of R&D for a new drug and growing demand towards greener products and processes. The cost of bringing a drug into the market has increased by more than a half over the last 10 years,while R&D returns have steadily declined since then. However, the cost of pharmaceutical development and manufacturing is important to ensure affordable access to medicine. In the view of these challenges and also initiatives from the drug regulatory mechanisms, the pharmaceutical industry has been prompted to seek new strategies and adopt new technologiesto replace the traditional energy and resource intensive routines without compromising the production quality. The aim of this project was to develop tools and methodologies to support and speed up in silico design and optimization efforts with a special focus on the crystallization process, which is a key separation and purification unit in the production process of high-purity pharmaceuticals and fine chemicals.

A mechanistic model library for multi-scale crystallization processes and experimental validation strategies were developed. The multi-scale models consisted of the compartmental models for the batch cooling and antisolvent crystallization processes at 218 L pilot scale.They were used to describe the dynamic changes in the crystal population (e.g. nucleationand growth) in combination with mass and energy balances as well as mixing information (momentum balance) for the suspension system. Crystallization of two well-known pharmaceutical compounds were utilized as the case studies. A steady-state compartmental model was developed for the batch cooling crystallization of acetaminophen (paracetamol), while a novel dynamic compartmental model to account for volume changes upon antisolvent filling was developed for the antisolvent crystallization of acetylsalicylic acid (aspirin). The prediction ability of the cooling crystallization model was tested against independent pilot scale experiments performed at 50 L scale for the model compound of paracetamol. A good agreement between experimental measurements and the simulation results was obtained in the absence of mixing failures such as particle settlement.

Global uncertainty and sensitivity analysis were applied to gain deeper process understandings and consequently to determine the strategies for the optimization of the process.The kinetic parameters and the operation parameters were considered as the sources of uncertainties or variations in the process. Their influence on the process output variance was quantified by uncertainty analysis. As an outcome, huge uncertainties on the process yield and final mean crystal size were obtained. Following the uncertainty analysis, the key parameters that have the highest influence on the process output (yield and final mean crystalsize) variance were identified by performing sensitivity analysis. The results emphasized that the process yield has the highest sensitivity to cooling/antisolvent feed time, while the final mean crystal size has the highest sensitivity to the mean of seed crystal size distribution under the studied conditions. The process risks were defined as failure to reach the target process specifications and its economic and environmental consequences for the given design space were quantified. Identification of the parameters having the highest influence provided a clear insight, on which parameters the optimization efforts should be focused on in order to reduce process output variance and mitigate associated risks.

Process analytical technology (PAT) tools offer valuable insight into the dynamics of the crystallization process, which are accessible online and in situ. Integration of PAT toolswith advanced control strategies provides great opportunities to reliably control the process dynamics, take fast actions in the presence of process disturbances, avoid deviations from product specifications and therefore reduce the batch-to-batch variations. With this purpose in mind, a data-driven control strategy was developed and applied experimentally on the batch cooling crystallization of ibuprofen. In this strategy, a radial basis functions (RBF) network model was trained in real time with experimental data obtained via PAT tools and a reference batch data. The aim was to optimize the cooling profile with the aid of trained RBF in orderto achieve the desired crystal population profile throughout the process. Comprehensive experiments, in which several process disturbances (initial supersaturation, impeller speed,water composition and seed size) existed, were performed to test the robustness of the proposed control strategy.

As a consequence, the data-driven control strategy could easilyhandle all disturbance scenarios in a way that the crystal population profile followed the desired reference successfully with less than 10 % offset in the most cases.The outcome of this project is an in silico tool that contains model-based and datadriven engineering approaches and tools with the purpose of aiding the robust and efficient process developments. The tool is expected to provide a rapid and cost efficient platform for generating, testing and evaluating different process design, optimization and control strategies for the crystallization process with a potential use in the pharmaceutical manufacturing.
Original languageEnglish
Place of PublicationKgs. Lyngby
PublisherTechnical University of Denmark
Number of pages238
Publication statusPublished - 2020

Cite this