Abstract Particle processes, such as crystallization, flocculation and emulsification constitute a large fraction of the industrial processes for removal of insolubles, product isolation, purification and polishing. The outcome of these processes typically needs to comply with a given set of quality attributes related to particle size, shape and/or yield. With recent technological advances in commercially available on-line/at-line particle analysis sensors, it is now possible to directly measure the particle attributes in real-time. This allows for developing new direct control strategies. In this work, a model predictive control (MPC) strategy is presented based on a hybrid machine-learning assisted particle model. The hybrid model uses mechanistic models for mass and population balances and machine learning for predicting the process kinetics. In the presented approach, the hybrid model is trained in real-time, during process operation. Combined with MPC, this allows for continuous refinement of the process model. Thereby, the calculated control actions are provided robustly. This approach can be employed with limited prior process knowledge, and allows for directly specifying the target product properties to the controller. The presented control strategy is demonstrated on a theoretical case of crystallization to show the potential of the presented methodology.
[1]
Zoltan K. Nagy,et al.
Recent advances in the monitoring, modelling and control of crystallization systems
,
2013
.
[2]
Krist V. Gernaey,et al.
Novel strategies for predictive particle monitoring and control using advanced image analysis
,
2019,
Computer Aided Chemical Engineering.
[3]
S. Y. Wong,et al.
Designing a lactose crystallization process based on dynamic metastable limit
,
2012
.
[4]
Zoltan K. Nagy,et al.
Development of a Model-Based Quality-by-Control Framework for Crystallization Design
,
2019
.