Efficient global sensitivity-based model calibration of a high-shear wet granulation process

• Published in: Chemical engineering science 2021-07, Vol.238, p.116569, Article 116569
• Main Authors: Yeardley, Aaron S., Bellinghausen, Stefan, Milton, Robert A., Litster, James D., Brown, Solomon F.
• Format: Article
• Language: English
• Subjects: Experimental design; Gaussian process; Global sensitivity analysis; Granulation; Model calibration; Sobol’ indices
• ISSN: 0009-2509; 1873-4405
• DOI: 10.1016/j.ces.2021.116569

•GSA used to identify important parameters in a wet granulation model case study.•Gaussian Process surrogate model utilised to calculate Sobol’ indices.•Modelling input-space reduced by 80% to four impactful parameters.•Model calibration workflow is proposed from identifying critical process parameters.•Enables a targeted experimental design that reduces experimental effort by 42.1%. Model-driven design requires a well-calibrated model and therefore needs efficient workflows to achieve this. This efficiency can be achieved with the identification of the critical process parameters (CPPs) and the most impactful modelling parameters followed by a targeted experimental campaign to prioritise the calibration of these. To identify these parameters it is essential to perform a global sensitivity analysis (GSA). Here, an efficient GSA is applied to a wet granulation case study with the Sobol’ indices used to identify the CPPs and impactful modelling parameters. The population balance, mechanistic model that is used requires considerable computational effort for a GSA so a Gaussian Process surrogate is utilised to interrogate the underlying model. These key results reduce the input-space by 80% enabling the proposal of a targeted experimental design and model calibration workflow. This substantially improves the ability to deploy model-based design to determine the impactful parameter values, reducing the experimental effort by 42.1% compared to a conventional experimental design.