(642h) An Efficient Data-Based Methodology to Identify the Design Space of Continuous Pharmaceutical Manufacturing Processes

Bhosekar, A., Rutgers, The State University of New Jersey
Ierapetritou, M., Rutgers, The State University of New Jersey
Metta, N., Rutgers, The State University of New Jersey
Ramachandran, R., Rutgers, The State University of New Jersey
Flowsheet modeling1,2 is an increasingly accepted tool to simulate and improve continuous processes for solid oral dosage manufacturing. Advances in continuous manufacturing have led to increased focus on developing accurate and detailed models for the unit operations. Flowsheet models developed using the unit models to simulate dynamic behavior of the continuous process are computationally expensive. Strategies are needed to overcome this limitation and fully realize the potential of these models. Specifically, the models can be utilized to enable decision making during detection of off-spec product, risk assessment and process optimization.

Tools such as sensitivity analysis are established3, and can be applied to flowsheet models to help identify the critical process parameters. While this helps in identifying variables where further research efforts can be focused on, the resulting problem is still high dimensional for implementing advanced analyses such as identification of design space. A design space is characterized by the range of variables within which the process satisfies equipment, quality and production constraints. Traditional methods to identify the design space require high sampling cost or closed form constraints, which limits its use with flowsheet models. Hence, effective strategies are required to handle problems typically encountered in pharmaceutical processes.

Current work uses surrogate based feasibility analysis method4 that builds surrogate models to approximate the feasibility function that characterizes the maximum constraint violation. This strategy uses a modified expected improvement function to identify samples close to the feasible region boundaries and unexplored regions. Previous work published used kriging4 and radial basis function5 as the surrogate models, which suffered limitations for high dimensional problems. Specifically for flowsheet models, the design space was identified using a set of two dimensional problems, which ignores interactions between the variables. In this work, a novel artificial neural network (ANN) based methodology is used to identify the design space. An ANN based surrogate model is built to approximate the feasibility function through carefully identifying samples using an adaptive sampling strategy. The unexplored regions are identified using the modified expected improvement function and a variance estimator. The variance of an unexplored sample is estimated using a statistical technique known as Jackknifing6,7. The developed approach is utilized to effectively identify the design space for an integrated direct compaction as well as wet granulation line that simulates a plant scale operation.


  1. Boukouvala F, Niotis V, Ramachandran R, Muzzio FJ, Ierapetritou MG. An integrated approach for dynamic flowsheet modeling and sensitivity analysis of a continuous tablet manufacturing process. Computers & Chemical Engineering. 2012;42:30-47.
  2. Yoon S, Galbraith S, Cha B, Liu H. Flowsheet modeling of a continuous direct compression process. 2018:121-139.
  3. Saltelli A, Annoni P, Azzini I, Campolongo F, Ratto M, Tarantola S. Variance based sensitivity analysis of model output. Design and estimator for the total sensitivity index. Computer Physics Communications. 2010;181(2):259-270.
  4. Boukouvala F, Ierapetritou MG. Derivative‐free optimization for expensive constrained problems using a novel expected improvement objective function. Aiche J. 2014;60(7):2462-2474.
  5. Wang Z, Ierapetritou M. A novel feasibility analysis method for black‐box processes using a radial basis function adaptive sampling approach. Aiche J. 2017;63(2):532-550.
  6. Eason J, Cremaschi S. Adaptive sequential sampling for surrogate model generation with artificial neural networks. Computers & Chemical Engineering. 2014;68:220-232.
  7. Jin Y, Li J, Du W, Qian F. Adaptive Sampling for Surrogate Modelling with Artificial Neural Network and its Application in an Industrial Cracking Furnace. The Canadian Journal of Chemical Engineering. 2016;94(2):262-272.